The DOI matches its Crossref record; 71 of the 88 references it cites were found in a registry or an index; 8 of the 27 sources its claims cite were checked for support.
Vision Transformers for Dense Prediction
Open accessAbstract
We introduce dense prediction transformers, an architecture that leverages vision transformers in place of convolutional networks as a backbone for dense prediction tasks. We assemble tokens from various stages of the vision transformer into image-like representations at various resolutions and progressively combine them into full-resolution predictions using a convolutional decoder. The transformer backbone processes representations at a constant and relatively high resolution and has a global receptive field at every stage. These properties allow the dense prediction transformer to provide finer-grained and more globally coherent predictions when compared to fully-convolutional networks. Our experiments show that this architecture yields substantial improvements on dense prediction tasks, especially when a large amount of training data is available. For monocular depth estimation, we observe an improvement of up to 28% in relative performance when compared to a state-of-the-art fully-convolutional network. When applied to semantic segmentation, dense prediction transformers set a new state of the art on ADE20K with 49.02% mIoU. We further show that the architecture can be fine-tuned on smaller datasets such as NYUv2, KITTI, and Pascal Context where it also sets the new state of the art. Our models are available at https://github.com/intel-isl/DPT.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
32 Own result · 23 Own interpretation · 11 No basis given · 10 No full text of the sources · 3 Supported · 2 Partly supported · 1 Supported so far · 1 Own method · accuracy not yet measured
Every source the paper cites for this claim was checked and supports it.
In the paper
“Specifically, we use the recently proposed vision transformer (ViT) [11] as a backbone architecture.”
§ Vision Transformers for Dense Prediction > 1. Introduction
What the sources say
- Supportedcites [11] 11. Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. source
Section 3.1 (p. 3) introduces and describes the Vision Transformer (ViT) architecture. The citation identifies the ViT used as the citing authors’ backbone; their choice to use it is not a claim attributed to this source. (LLM, conf=0.93)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Virtually all existing architectures for dense prediction are based on convolutional networks [6,31,34,42,49,50,53].”
§ Vision Transformers for Dense Prediction > 1. Introduction
What the sources say
- Partly supportedcites [6] 6. Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L. Yuille. DeepLab: Semantic im- age segmentation with deep convolutional nets, atrous con- volution, and fully connected crfs. TPAMI, 40(4):834-848, 2018. sourceExperimental — accuracy not yet measured
The Introduction (p. 1) says convolutional models such as PixelCNN had received much more attention for pixel-by-pixel image generation. It does not establish that virtually all architectures across the broader category of dense prediction are convolutional; it also discusses recurrent models and proposes a self-attention alternative (Introduction, p. 2). (LLM, conf=0.78)
- Not checked — no full text of the sourcecites [31] 31. Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- Net: Convolutional networks for biomedical image segmen- tation. In MICCAI, 2015. source
- Supportedcites [34] 34. Evan Shelhamer, Jonathan Long, and Trevor Darrell. Fully convolutional networks for semantic segmentation. CVPR, 2015. source
The paper is an example of a convolutional-network architecture for dense prediction: its Abstract (p. 1) describes predicting a depth map, and §3.1.2 (p. 3) identifies its convolutional layers and target-depth output. As one of several citations, it supports the sentence by exemplifying such an architecture, rather than establishing the prevalence claim on its own. (LLM, conf=0.88)
- Could not verifycites [42] 42. Jingdong Wang, Ke Sun, Tianheng Cheng, Borui Jiang, Chaorui Deng, Yang Zhao, Dong Liu, Yadong Mu, Mingkui Tan, Xinggang Wang, Wenyu Liu, and Bin Xiao. Deep high-resolution representation learning for visual recogni- tion. TPAMI, 2020. sourceExperimental — accuracy not yet measured
Section 2.1 (p. 8) describes transformer language-model architectures, not architectures for dense prediction. Searches for dense prediction, convolutional networks, and pixel-wise segmentation found no passage addressing whether dense-prediction architectures are predominantly convolutional. (LLM, conf=0.78)
- Not matchedcites [49] 49. Fisher Yu and Vladlen Koltun. Multi-scale context aggrega- tion by dilated convolutions. In ICLR, 2016.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked — no full text of the sourcecites [50] 50. Yuhui Yuan, Xilin Chen, and Jingdong Wang. Object- contextual representations for semantic segmentation. In ECCV, 2020. source
- Not checked — no full text of the sourcecites [53] 53. Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In CVPR, 2017. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“We notice that the biggest gains in performance for zero-shot transfer were achieved for datasets that feature dense, high-resolution evaluations [15,55,59].”
§ Vision Transformers for Dense Prediction > Additional information
What the sources say
- Not checked — no full text of the sourcecites [15] 15. Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? The KITTI vision benchmark suite. In CVPR, 2012. source
- Supportedcites [55] 55. D. J. Butler, J. Wulff, G. B. Stanley, and M. J. Black. A naturalistic open source movie for optical flow evaluation. In ECCV, 2012. source
The citation identifies MPI-Sintel as one of the datasets in the citing authors’ comparison; the performance gains are the citing authors’ observation, not a result attributed to this paper. Design Decisions and Comparison to Middlebury (p. 5) describes Sintel’s higher-resolution frames, and The Sintel Data Set (p. 9) describes its flow fields and pixel-level error evaluation. (LLM, conf=0.86)
- Not checked — no full text of the sourcecites [59] 59. Thomas Schöps, Johannes L. Schönberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and An- dreas Geiger. A multi-view stereo benchmark with high- resolution images and multi-camera videos. In CVPR, 2017. source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Note that ResNext101-WSL was pretrained on a billion-scale corpus of weakly supervised data [25] in addition to ImageNet pretraining.”
§ Vision Transformers for Dense Prediction > Skip connections.
What the sources say
- Partly supportedcites [25] 25. Dhruv Mahajan, Ross Girshick, Vignesh Ramanathan, Kaiming He, Manohar Paluri, Yixuan Li, Ashwin Bharambe, and Laurens van der Maaten. Exploring the limits of weakly supervised pretraining. In ECCV, 2018. sourceExperimental — accuracy not yet measured
Section 2.3 (p. 4) identifies the models as ResNeXt-101, and Section A.5 (p. 23) says they were pretrained on billions of Instagram images using hashtags as labels. But it describes subsequent training on ImageNet as fine-tuning, not additional ImageNet pretraining; Section 3.2 (p. 12) also distinguishes Instagram-pretrained models from ImageNet-pretrained baselines. (LLM, conf=0.88)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While ignoring the token yields good performance, projection provides slightly better performance on average.”
§ Vision Transformers for Dense Prediction > Skip connections.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Adding the token, on the other hand, yields worse performance than simply ignoring it.”
§ Vision Transformers for Dense Prediction > Skip connections.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Both DPT variants significantly outperform the state of the art.”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Note that both architectures have similar latency to MiDaS (Table 9).”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“ViT-Large outperforms all other backbones but is also almost three times larger than ViT-Base and ViT-Hybrid.”
§ Vision Transformers for Dense Prediction > Skip connections.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We again see that DPT can provide strong performance even on smaller datasets.”
§ Vision Transformers for Dense Prediction > ADE20K.
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Probably not a substantial claim. 11
- It may state two separate claims as one. 13
- The quote only partly states this claim. 1
Every source the paper cites for this claim was checked and supports it.
In the paper
“In particular, it has recently been demonstrated that a direct application of token-based transformer architectures that have been successful in NLP can yield competitive performance on image classification [11].”
§ Vision Transformers for Dense Prediction > 2. Related Work
What the sources say
- Supportedcites [11] 11. Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. source
The Introduction (p. 1) says image patches are treated like NLP tokens and fed to a standard Transformer. The Conclusion (p. 9) reports that this approach matches or exceeds the state of the art on many image-classification datasets when coupled with pre-training on large datasets. (LLM, conf=0.93)
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“A key insight of this work was that, like transformer models in NLP, vision transformers need to be paired with a sufficient amount of training data to realize their potential.”
§ Vision Transformers for Dense Prediction > 2. Related Work
What the sources say
- Supportedcites [11] 11. Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. source
The Introduction (p. 1) says ViT performs modestly on mid-sized datasets and does not generalize well with insufficient data, but attains excellent results when pre-trained at sufficient scale. Section 4.3 (p. 6) reports that larger ViT models show their full benefit only with the largest pre-training dataset tested. The Conclusion (p. 9) also describes the approach as working well when coupled with pre-training on large datasets. (LLM, conf=0.94)
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Supported
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Similarly, the ViT-Hybrid architecture extracts features at 1 16 the input resolution, which is twice as high as the lowest-resolution features that are commonly used with convolutional backbones.”
§ Vision Transformers for Dense Prediction > 3. Architecture
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“It contains about 1.4 million images and is, to the best of our knowledge, the largest training set for monocular depth estimation that has ever been compiled.”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When applied to semantic segmentation, dense vision transformers set a new state of the art on ADE20K with 49.02% mIoU.”
§ Vision Transformers for Dense Prediction > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We observe that the DPT tends to produce cleaner and finer-grained delineations of object boundaries and that the predictions are also in some cases less cluttered.”
§ Vision Transformers for Dense Prediction > ADE20K.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“DPT can better reconstruct fine details while also improving global coherence in areas that are challenging for the convolutional architecture (for example, large homogeneous regions or relative depth arrangement across the image).”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The average relative improvement over the best published architecture, MiDaS, is more than 23% for DPT-Hybrid and 28% for DPT-Large.”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“DPT-Hybrid achieves this with a comparable network capacity (Table 9), while DPT-Large is about 3 times larger than MiDaS.”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We observe the tendency that attention is spatially more localized close to the reference token in shallow layers (leftmost columns), whereas deeper layers (rightmost columns) frequently attend across the whole image.”
§ Vision Transformers for Dense Prediction > Additional information
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“ViT-Base has comparable performance to ResNext101-WSL, while ViT-Hybrid and ViT-Large improve performance even though they have been pretrained on significantly less data.”
§ Vision Transformers for Dense Prediction > Skip connections.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“While we observe a general trend of an improvement in per-class IoU in comparison to the baseline [51], we do not observe a strong pattern across classes.”
§ Vision Transformers for Dense Prediction > Additional information
What the sources say
- Not checked — no full text of the sourcecites [51] 51. Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Zhi Zhang, Haibin Lin, Yue Sun, Tong He, Jonas Muller, R. Manmatha, Mu Li, and Alexander Smola. ResNeSt: Split- attention networks. arXiv preprint arXiv:2004.08955, 2020.
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“While we observe a general trend of an improvement in per-class IoU in comparison to the baseline [51], we do not observe a strong pattern across classes.”
§ Vision Transformers for Dense Prediction > Additional information
What the sources say
- Not checked — no full text of the sourcecites [51] 51. Hang Zhang, Chongruo Wu, Zhongyue Zhang, Yi Zhu, Zhi Zhang, Haibin Lin, Yue Sun, Tong He, Jonas Muller, R. Manmatha, Mu Li, and Alexander Smola. ResNeSt: Split- attention networks. arXiv preprint arXiv:2004.08955, 2020.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We observe more details and also better global depth arrangement in DPT predictions when compared to the fullyconvolutional baseline.”
§ Vision Transformers for Dense Prediction > Additional information
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The transformer backbone processes representations at a constant and relatively high resolution and has a global receptive field at every stage.”
§ Vision Transformers for Dense Prediction > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As long as the image size is divisible by p, the embedding procedure can be applied and will produce a varying number of image tokens N p .”
§ Vision Transformers for Dense Prediction > 3. Architecture
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Unlike fully-convolutional networks, the vision transformer backbone foregoes explicit downsampling operations after an initial image embedding has been computed and maintains a representation with constant dimensionality throughout all processing stages.”
§ Vision Transformers for Dense Prediction > 1. Introduction
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Since transformers are known to realize their full potential only when an abundance of training data is available, monocular depth estimation is an ideal task to test the capabilities of DPT.”
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This is in stark contrast to convolutional networks, which progressively increase their receptive field as features pass through consecutive convolution and downsampling layers.”
§ Vision Transformers for Dense Prediction > 3. Architecture
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“DPT-Large performs slightly worse, likely because of the significantly smaller dataset compared to our previous experiments.”
§ Vision Transformers for Dense Prediction > ADE20K.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“DeIT trains the ViT architecture with a more data-efficient pretraining procedure.”
§ Vision Transformers for Dense Prediction > Skip connections.
What the sources say
- Not checked — no full text of the sourcecites [38] 38. Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through at- tention. arXiv preprint arXiv:2012.12877, 2020. source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“Note that the DeIT-Base architecture is identical to ViT-Base, while DeIT-Base-Dist introduces an additional distillation token, which we ignore in the Reassemble operation.”
§ Vision Transformers for Dense Prediction > Skip connections.
What the sources say
- Not checked — no full text of the sourcecites [38] 38. Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through at- tention. arXiv preprint arXiv:2012.12877, 2020. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Similar to prior work on transformers, DPT unfolds its full potential when trained on large-scale datasets.”
§ Vision Transformers for Dense Prediction > 5. Conclusion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Note that results for DPT and Mi-DaS are computed at the same input resolution (384 pixels).”
§ Vision Transformers for Dense Prediction > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper relies on this without support.
The choice of backbone architecture has a large influence on the capabilities of the overall model, because information lost in the encoder cannot be recovered in the decoder.
“However, it is widely recognized that the choice of backbone architecture has a large influence on the capabilities of the overall model, as any information that is lost in the encoder is impossible to recover in the decoder.”
The paper relies on this without support.
Feature resolution and granularity are critical for dense prediction, where the architecture should ideally resolve features at or close to the input resolution.
“While feature resolution and granularity may not matter for some tasks, such as image classification, they are critical for dense prediction, where the architecture should ideally be able to resolve features at or close to the resolution of the input image.”
The paper relies on this without support.
Convolutions are linear operators with a limited receptive field.
“Convolutions, by definition, are linear operators that have a limited receptive field.”
The paper relies on this without support.
Downsampling intermediate representations is necessary to keep memory consumption at levels feasible with existing computer architectures.
“Downsampling the intermediate representations is necessary to keep memory consumption at levels that are feasible with existing computer architectures.”
The paper relies on this without support.
The transformer backbone maintains a constant feature resolution.
“Since the transformer backbone maintains a constant feature resolution, it is not clear at which points in the backbone features should be tapped.”
The paper relies on this without support.
Performance of fully-convolutional architectures suffers heavily when performing inference at an input resolution significantly different from the training resolution.
“Performance thus suffers heavily when performing inference at an input resolution that is significantly different from the training resolution.”
The paper relies on this without support.
Transformer encoders have a global receptive field in every layer.
“Transformer encoders, on the other hand, have a global receptive field in every layer.”
The paper relies on this without support.
The reduced meta-dataset datasets provide high-quality ground truth.
“We choose these datasets since they provide high-quality ground truth.”
The paper relies on this without support.
Convolutional architectures offer natural points of interest for passing features from the encoder to the decoder, namely before or after downsampling of the representation.
“Convolutional architectures offer natural points of interest for passing features from the encoder to the decoder, namely before or after downsampling of the representation.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Vision transformers, which are a deep learning approach to encoding low-dimensional input into a high dimensional feature space, have established new frontiers in depth estimation compared to convolutional neural networks (Ranftl et al., 2021).”
From Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on aerial lidar (2024) · cites this paper as Ranftl et al., 2021
In the citing paper
“However, the Transformer framework does not involve dimensionality reduction10,12, thus keeping all attention layer traceable to the original input features13, thus making the models interpretable.”
From Transformer for one stop interpretable cell type annotation (2023) · cites this paper as [10]
In the citing paper
“For example, the change of dimensions and non-linear aggregation of features throughout the autoencoder’s deep processing stages leads to untraceable and uninterpretable latent space and loss of information and feature resolution9,10.”
From Transformer for one stop interpretable cell type annotation (2023) · cites this paper as [10]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Vision Transformers for Dense Prediction > 4.1. Monocular Depth Estimation
“Our architecture matches or improves state-of-the-art performance on both datasets in all metrics.”
§ Vision Transformers for Dense Prediction > 5. Conclusion
“| BTS [20] | 0.956 | 0.993 | 0.998 | 0.059 2.756 | 0.096 | | DPT-Hybrid 0.959 | 0.995 | 0.999 | 0.062 2.573 | 0.092 |”
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 4×
“We perform a similar experiment with the hybrid architecture in Table 6 (bottom), where R0 and R1 refer to us-ing features from the first and second downsampling stages of the ResNet50 embedding network.”
MethodInfluentialVaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
U-Net: Convolutional Networks for Biomedical Image Segmentation
Ronneberger, O., Fischer, P., & Brox, T. (2015) · Lecture notes in computer science · doi:10.1007/978-3-319-24574-4_28 · Cited by 99,404 · cited 3×
“These include training at higher input resolution (if the computational budget permits), dilated convolutions [51] to rapidly increase the receptive field without downsampling, appropriately-placed skip connections from multiple stages of the encoder to the decoder [33], or, more recently, by connecting multiresolutio…”
BackgroundAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
ImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 5×
“The encoder is frequently based on an image classification network, also called the backbone, that is pretrained on a large corpus such as ImageNet [9].”
InfluentialBERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 6×
“Following work in NLP, the ViT additionally adds a special token that is not grounded in the input image and serves as the final, global image representation which is used for classification.”
BackgroundMethodInfluentialAn Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al. (2020) · arXiv · doi:10.48550/arxiv.2010.11929 · Cited by 47,472 · cited 8×
“On a high level, the vision transformer (ViT) [11] operates on a bag-of-words representation of the image [38].”
BackgroundMethodResultInfluentialFully convolutional networks for semantic segmentation
Long, J., Shelhamer, E., & Darrell, T. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298965 · Cited by 38,834 · cited 2×
“Virtually all existing architectures for dense prediction are based on convolutional networks [6, 33, 36, 44, 51, 52, 55].”
BackgroundLanguage Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
Research works printed but not linked: 3
- yuan, 2020, Objectcontextual representations for semantic segmentation, ECCV
- godard, 2017, Unsupervised monocular depth estimation with leftright consistency, CVPR
- casser, 2019, Unsupervised learning of depth and ego-motion: A structured approach, AAAI
Cited by
Sorted by citations
Items 1,101–1,200 of 2,248
Ren, J., Xiang, M., Zhu, J., et al. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.02689 · Cited by 3
“We use the DPT head [47] to predict the Gaussian parameters G 1 , 2 for each view.”
Domain Shared and Specific Prompt Learning for Incremental Monocular Depth Estimation
Yang, Z., Li, L., Zhang, J., Wang, T., Sun, Y., & Yan, C. C. (2024) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3664647.3681155 · Cited by 3
“To address the aforementioned problems, some approaches [32, 33] attempt to reduce the domain gaps by employing multi-domain learning across as many domains as possible and alleviate the influence of the depth space variations by predicting relative depth maps.”
Gated Fields: Learning Scene Reconstruction from Gated Videos
Ramazzina, A., Walz, S., Dahal, P., Bijelic, M., & Heide, F. (2024) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52733.2024.01002 · Cited by 3
“We compare our method against 9 feed-forward depth estimation methods, namely SimIPU [53], AdaBins [10], DPT [70], DepthFormer [54] and CREStereo [50] for monocular and stereo RGB methods, Gated2Gated [89] and Gat-edStereo [90] for monocular and stereo gated estimation methods.”
MethodWu, M., Qi, H., Fan, W., et al. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10447105 · Cited by 3
“We compare our CHAT method with state-of-the-art methods: (1) DPT-Hybrid [17], (2) EndoSFM [10], (3) Table 1 : Six metrics are used for evaluation ( P k and G k are predicted and ground truth depths at point k , respectively).”
MethodIyer, H., Eiris, R., & Jeong, H. (2025) · IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) · doi:10.1109/vrw66409.2025.00218 · Cited by 3
“Digital twin made from a prelim-inary analysis using MiDaS [32, 33] and OpenCV [4] is shown in Figure 3.”
Dual-Encoder framework for multi-abnormality classification of VinDr-CXR dataset
Arora, R., Saini, I., & Sood, N. (2023) · International Conference on Secure Cyber Computing and Communication (ICSCCC) · doi:10.1109/icsccc58608.2023.10176862 · Cited by 3
“The main contribution of this paper as follows: • We employed dual-encoder framework with Vision Transformer and MobileNet encoders [5-6].”
MethodJung, C. R., Lee, J., Kim, G., Kim, J., Park, S., & Cha, H. (2025) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3711875.3729161 · Cited by 8
“VFMs used DINOv2 [36] as a visual feature encoder, with DPT [41] and Mask2Former [6] as decoders to produce final results for MDE and SS, respectively.”
Enhanced Feature Pyramid Vision Transformer for Semantic Segmentation on Thailand Landsat-8 Corpus
Intarat, K., Rakwatin, P., & Panboonyuen, T. (2022) · Information · doi:10.3390/info13050259 · Cited by 3
Real-Time Interaction for 3D Pixel Human in Virtual Environment
Deng, H., Zhang, Q., Jin, H., et al. (2023) · Applied Sciences · doi:10.3390/app13020966 · Cited by 3
Zhu, J., Wang, H., & Shi, M. (2024) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2024.3428487 · Cited by 3
“Compared with the 2D coordinates in images, there is an additional dimension measuring the distance from the object to the image plane in the 3D coordinates, which can be converted to image depth [14].”
BackgroundMethodForce-EvT: A Closer Look at Robotic Gripper Force Measurement with Event-Based Vision Transformer
Guo, Q., Yu, Z., Fu, J., et al. (2024) · International Conference on Reconfigurable Mechanisms and Robots (ReMAR) · doi:10.1109/remar61031.2024.10617687 · Cited by 4
“For instance, Ranftl et al. presented a novel architecture called dense prediction transformers, which employed Vision Transformer instead of convolutional networks as the foundational structure for tasks requiring dense predictions [28].”
BackgroundReactive Collision Avoidance for Safe Agile Navigation
Saviolo, A., Picello, N., Mao, J., et al. (2025) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra55743.2025.11127284 · Cited by 3
ADC-GS: Pose-Free 3D Gaussian Splatting with Adaptive Depth Consistency
Liu, R., Shen, S., Jiang, N., et al. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10890011 · Cited by 3
“Given a set of images { I i } =1 and mono-depth maps { D i } =1 generated by an off-the-shelf monocular depth network, e.g., DPT [31].”
Edge Enhanced Scene Understanding and Depth Estimation
Islam, N. U., & Shah, S. H. (2025) · International Conference on Image, Video and Signal Processing (IVSP) · doi:10.1145/3749859.3749861 · Cited by 3
“In ZoeDepth, the DPT encoder-decoder architecture [22] is employed as a foundation model.”
Multi-Resolution Monocular Depth Map Fusion by Self-Supervised Gradient-Based Composition
Dai, Y., Yi, R., Zhu, C., He, H., & Xu, K. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i1.25123 · Cited by 3
Optimizing Local-Global Dependencies for Accurate 3D Human Pose Estimation
Xu, G., Zhang, G., Ye, L., et al. (2025) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2025.3585610 · Cited by 3
“Compared to CNNs, the success of Vision Transform-ers (ViTs) [40], [41] in image classification mainly stems from their ability to model long-range dependencies in the input images [18], [42], [43], [44].”
TSUDepth: Exploring temporal symmetry-based uncertainty for unsupervised monocular depth estimation
Zhu, Y., Ren, R., Dong, W., Li, X., & Shi, G. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.128165 · Cited by 4
“We adopt the DPT-Hybrid in Dense Prediction Transformer (DPT) [12] as our backbone, which successfully integrates the Vision Transformer (ViT) structure into dense prediction tasks.”
MethodInfluentialComparison of CNN and Vision Transformers for Wildfire Detection: A Proxy for Stubble Burning
Choudhry, A., Jain, C., Singh, S., et al. (2025) · International Conference on Disruptive Technologies (ICDT) · doi:10.1109/icdt63985.2025.10986498 · Cited by 3
Türe, M., Çıklabakkal, M. E., Erdem, A., et al. (2021) · Computer Graphics Forum · doi:10.1111/cgf.14392 · Cited by 3
Park, J., Kim, S., Park, W.-H., Song, S., & Yoo, H. (2024) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2024.3446662 · Cited by 3
“[10] Fig.”
DGTR: Distributed Gaussian Turbo-Reconstruction for Sparse-View Vast Scenes
Li, H., Gao, Y., Peng, H., Wu, C., Ye, W., Zhan, Y., Zhao, C., Zhang, D., Wang, J., & Han, J. (2025) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra55743.2025.11127313 · Cited by 3
“…correlation is used as loss L reg to measures the distribution difference between 2D depth maps: where D is the depth map rendered by Gaussian model, D est is the depth maps predicted by pre-trained depth estimator DPT [40], Cov( · , · ) and Var( · , · ) denote covariance and variance separately.”
Pseudo-LiDAR Meets Agriculture: Leveraging 3D Monocular Point Cloud Processing for Coffee Beans
Febriana, A., Farady, I., Lin, P.-C., Lin, C., & Muchtar, K. (2023) · International Conference on Computer System, Information Technology, and Electrical Engineering (COSITE) · doi:10.1109/cosite60233.2023.10249478 · Cited by 3
“We used MiDaS Small-CNN, which is known for its computational efficiency, and two variants of the dense prediction transformer (DPT) [16] called DPT-Hybrid and DPT-Large.”
MethodA Transformer-Based Image-Guided Depth-Completion Model with Dual-Attention Fusion Module
Wang, S., Jiang, F., & Gong, X. (2024) · Sensors · doi:10.3390/s24196270 · Cited by 3
“While the plain ViT [22,39] divides the image into patches only once, leading to single-scale features with lower resolution, our model enhances this approach.”
BackgroundMethodInfluentialChallenging Weather for License Plate Detection: Impaired visibility
Ibrahim, M. A., El-Saady, K., & Kenk, M. A. (2023) · Novel Intelligent and Leading Emerging Sciences Conference (NILES) · doi:10.1109/niles59815.2023.10296738 · Cited by 3
Semantic RGB-D Image Synthesis
Li, S., Li, R., & Gall, J. (2023) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw60793.2023.00101 · Cited by 3
“Comparison of the real depth image (a) with the depth map that is estimated by [21] from the real color image (b) and the depth image that is generated by our method from the label map (c).”
MethodInfluentialA Video Shot Occlusion Detection Algorithm Based on the Abnormal Fluctuation of Depth Information
Liao, J., Duan, H., Zhao, W., Feng, K., Yang, Y., & Chen, L. (2023) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2023.3295243 · Cited by 3
“Examples of depth maps predicted by different depth estimation methods [8], [55], [56], [57].”
BackgroundInfluentialBamani, E., Nissinman, E., Koenigsberg, L., et al. (2025) · IEEE Access · doi:10.1109/access.2025.3616050 · Cited by 3
“Therefore, we incorporate the MiDaS Monocular Depth Estimation model [56], [57] which adds estimated depth information to the observed RGB image.”
Xu, C., Han, C., Yang, H., et al. (2024) · Journal of King Saud University - Computer and Information Sciences · doi:10.1016/j.jksuci.2024.102222 · Cited by 3
RD-NERF: Neural Robust Distilled Feature Fields for Sparse-View Scene Segmentation
Ma, Y., Dou, B., Zhang, T., et al. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10447068 · Cited by 3
“These methods integrate pre-trained DINO-ViT[18] for semantic distillation and incorporate the learning of sparse user-provided annotations with the learned 3D feature volume in order to achieve object segmentation in 3D scenes.”
MethodGraph-Based Similarity of Neural Network RepresentationsPreprint
Chen, Z., Lu, Y., Hu, J., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.11165 · Cited by 3
“It is indisputable that Deep Neural Network (DNN) has achieved the most advanced performance in the fields like computer vision [56], natural language processing [10], etc.”
Fast Depth Estimation of Object via Neural Network Perspective Projection
Han, Y., Chen, Y., Li, H., et al. (2022) · IEEE Data Driven Control and Learning Systems Conference (DDCLS) · doi:10.1109/ddcls55054.2022.9858358 · Cited by 3
“At present, there are many supervised learning methods: different loss functions-based methods [18], conditional random fields-based meth-ods [19], adversarial learning-based methods [20], and trans-former structure-based methods [21].”
MethodSpherical World-Locking for Audio-Visual Localization in Egocentric Videos
Yun, H., Gao, R., Ananthabhotla, I., Kumar, A. K., Donley, J., Li, C., Kim, G., Ithapu, V. K., & Murdock, C. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-72691-0_15 · Cited by 3
Wu, D., Ma, Z., Xu, W., He, H., & Li, Z. (2024) · World Electric Vehicle Journal · doi:10.3390/wevj15030123 · Cited by 3
“Selecting the appropriate scale parameter s and the number of pyramid layers N in OPENCV can reduce the computation time and the mismatch rate of ORB features during usage [18].”
BackgroundA Context-Aware Doorway Alignment and Depth Estimation Algorithm for Assistive Wheelchairs
Tennekoon, S., Wedasingha, N. D., Welhenge, A. M., et al. (2025) · Computers · doi:10.3390/computers14070284 · Cited by 3
Gafurov, A., & Иванов, М. А. (2024) · Land · doi:10.3390/land13122054 · Cited by 3
“With the advent of Vision Transformers (ViT), new ways of processing images that do not depend solely on local convolutions have been proposed [14,15].”
Learning-Based 3D Reconstruction Methods for Non-Collaborative Surfaces—A Metrological Evaluation
Yan, Z., Padkan, N., Trybała, P., Farella, E. M., & Fabio Remondino, A. (2025) · Metrology · doi:10.3390/metrology5020020 · Cited by 6
Sparse depth densification for monocular depth estimation
Liang, Z., Fang, T., Hu, Y., et al. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-15757-4 · Cited by 3
“Among these compared methods, some methods gain high depth estimation accuracy by using additional training data [8, 23, 25, 48], several recent methods [3, 36, 53] apply the transformer framework to the depth estimation.”
Channel-Spatial Mutual Attention Network for 360° Salient Object Detection
Zhang, Y., Hamidouche, W., & Déforges, O. (2022) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr56361.2022.9956354 · Cited by 3
“ER = the hybrid-ViT-based encoder [28] for ER image.”
BackgroundSemicalibrated Relative Pose from an Affine Correspondence and Monodepth
Hruby, P., Pollefeys, M., & Baráth, D. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-73661-2_3 · Cited by 3
Using full-scale feature fusion for self-supervised indoor depth estimation
Cheng, D., Chen, J., Lv, C., et al. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-16581-6 · Cited by 3
“In the supervised method, the representative DPT [23], whose model has 123.15 M parameters and 82.49G FLOPs.”
InfluentialLightweight Multi-Frequency Enhancement Network for RGB-D Video Salient Object Detection
Suolang, D., He, J., Tsering, W., Fu, K., Li, X., & Zhao, Q. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10890388 · Cited by 3
“For more comparison, we use DPT [40] to generate synthetic depth maps for VSOD benchmark datasets (DAVIS [41], DAVSOD [12]) and conduct experiments on them, as well as on RGB-D SOD datasets (NJU2K [42], NLPR [43]).”
Aggregation of Masked Outputs for Improving Accuracy–Cost Trade-Off in Semantic Segmentation
Suh, M.-K., & Seo, S. (2023) · IEEE Access · doi:10.1109/access.2023.3265077 · Cited by 2
MonoPlane: Exploiting Monocular Geometric Cues for Generalizable 3D Plane Reconstruction
Wang, Z., Liu, J., Zhang, S., et al. (2024) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros58592.2024.10802672 · Cited by 3
“DPT [41] proposes a network which combines the advantage of vision transformers [42] and CNNs to achieve global receptive field while preserving fine-grained details on depth estimation.”
MonoDGAE: depth-guided attention and bilateral filtering for robust monocular 3D object detection
Bitwire, G. A., Kakuba, S., Cha, D. W., & Han, D. S. (2025) · Artificial Life and Robotics · doi:10.1007/s10015-025-01048-8 · Cited by 3
“With the rise of deep learning, supervised approaches like MiDaS [12] and DPT [13] introduced robust depth prediction by leveraging multi-scale learning and transformers, respectively.”
Can large language models understand depth in monocular images without prior vision knowledge?
Xia, Z., Wu, T., & Wu, B. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.111680 · Cited by 7
AI-Powered Visual Sensors and Sensing: Where We Are and Where We Are Going
Nguyen, H., Vo, M., Hyatt, J. S., & Wang, Z. (2025) · Sensors · doi:10.3390/s25061758 · Cited by 5
TSGaussian: Semantic and depth-guided Target-Specific Gaussian Splatting from sparse views
Zhao, L., Bao, Z., Xie, Y. C., et al. (2025) · Image and Vision Computing · doi:10.1016/j.imavis.2025.105706 · Cited by 3
DepthNet: A Monocular Depth Estimation Framework
Anunay, A., Pankaj, & Dhiman, C. (2021) · International Conference on Engineering and Emerging Technologies (ICEET) · doi:10.1109/iceet53442.2021.9659770 · Cited by 3
一种融合视觉Transformer和扩散模型的单视点内窥镜手术光场重建方法(特邀)
Chenming, 韩. H., & Gaochang, 吴. W. (2024) · Laser & Optoelectronics Progress · doi:10.3788/lop241272 · Cited by 3
Towards Robust Monocular Depth Estimation in Non-lambertian Surfaces
Zhang, J., Li, J., Huang, Y., Wang, Y., Zheng, J., Shen, L., & Cao, Z. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-91569-7_12 · Cited by 3
STOPNet: Multiview-based 6-DoF Suction Detection for Transparent Objects on Production Lines
Kuang, Y., Han, Q., Li, D., et al. (2024) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra57147.2024.10611585 · Cited by 3
“Dense Prediction Transformers, i.e. DPT [23] is a large-scale vision transformer that can be applied to many dense image-to-image prediction tasks, such as semantic segmentation, monocular depth estimation, etc .”
BackgroundMethodInfluentialGuiding Local Feature Matching with Surface Curvature
Wang, S., Kannala, J., Pollefeys, M., et al. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01648 · Cited by 3
“To demonstrate this, we fine-tune the state-of-the-art MiDaS [42] depth predictor to increase the feature matching accuracy.”
BackgroundMethodInfluentialSphereDepth: Panorama Depth Estimation from Spherical Domain
Yan, Q., Wang, Q., Zhao, K., Li, B., Chu, X., & Deng, F. (2022) · International Conference on 3D Vision (3DV) · doi:10.1109/3dv57658.2022.00022 · Cited by 4
“Meanwhile, Adabins [3] uses ViT[10] to guess the best discrete depth distribution, and DPT [32] directly proposes a transformer network that can perform dense prediction.”
MethodBenchmarking Monocular Depth Estimation Models for VR Content Creation from a User Perspective
Dickson, A., Knott, A., & Zollmann, S. (2021) · International Conference on Image and Vision Computing New Zealand (IVCNZ) · doi:10.1109/ivcnz54163.2021.9653344 · Cited by 3
Lu, Y., Wang, H., Ning, Q., Liu, Z., Zang, Y., Liao, Z., & Yan, Z. (2026) · Sensors · doi:10.3390/s26072026 · Cited by 3
“…feature interaction, the model outputs geometric factors in parallel through distinct prediction heads: a Dense Prediction Transformer (DPT) [30] head predicts dense ray directions and depths; a Pose head predicts quaternions and translations; and a Multi-Layer Perceptron (MLP) [31] head…”
Fan, W., Jiang, W., Shi, H., Zeng, H., Chen, Y., & Luó, X. (2024) · IEEE Transactions on Medical Robotics and Bionics · doi:10.1109/tmrb.2024.3407384 · Cited by 4
“Ranftl et al. [8] and Bae et al. [27] proposed transformer architectures (DPT and MonoFormer) for monocular dense depth prediction and achieved good effectiveness especially in the depth structure prediction.”
BackgroundMethodResultInfluentialMonocular Depth Estimation Based on Residual Pooling and Global-Local Feature Fusion
Li, L., Zheng-you, L., Liang, X., & Li, S. (2024) · IEEE Access · doi:10.1109/access.2024.3453942 · Cited by 4
“Ranftl et al. [23] utilized a network architecture based on Vision Transformer-CNN as an encoder, providing more fine-grained and globally consistent predictions, but the model training requires large datasets and takes longer training times.”
BackgroundMethodEndoDAV: Depth Any Video in Endoscopy with Spatiotemporal Accuracy
Zhou, Z., Yang, C., Yang, P., Yang, X., & Shen, W. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-05114-1_19 · Cited by 3
Sensing Range Extension for Short-Baseline Stereo Camera Using Monocular Depth Estimation
Seo, B.-S., Park, B., & Choi, H. Y. (2022) · Sensors · doi:10.3390/s22124605 · Cited by 3
Wang, T., Lu, J., Wan, B., et al. (2026) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2026.114489 · Cited by 2
Streamed Optical Flow Adaptation From Synthetic to Real Dental Scenes
Lewandowska, E., Jezierska, A., & Węsierski, D. (2025) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2025.3573265 · Cited by 2
“Using an expectation-maximization optimization framework, the pseudo-flow labels were estimated with RAFT [22], while images were generated from the depth maps [38].”
Dutta, S., Vasim, A., Rezatofighi, H., et al. (2026) · ACM Transactions on Multimedia Computing Communications and Applications · doi:10.1145/3787522 · Cited by 2
“Unlike the traditional setting, the set of classes seen during training, 𝒞 train , may differ from the set of classes encountered during inference, 𝒞 test , under the open-vocabulary setup.”
Prakash, S., Walton, D. R., dos Anjos, R. K., Steed, A. J., & Ritschel, T. (2025) · IEEE Transactions on Visualization and Computer Graphics · doi:10.1109/tvcg.2025.3549541 · Cited by 2
“[29,30] focus especially in the AR use case.”
Dense Depth Map Estimation Based on Camera–LiDAR Sensor Fusion
Bong, E.-J., & Kee, S.-C. (2026) · IEEE Sensors Journal · doi:10.1109/jsen.2025.3649237 · Cited by 2
“Hybrid methods that combine both architectures have also been explored [19], [20].”
SVLGaussian: Single View Language Gaussian Splatting
Wang, J., Ding, M., Tan, M., Zhang, L., Fan, J., Pan, W., Yu, Z., & Ding, J. (2026) · CAAI Transactions on Intelligence Technology · doi:10.1049/cit2.70148 · Cited by 2
Amuche, C. I., Zhang, X., Ukwuoma, C. C., et al. (2025) · International Conference on Software Engineering and Computer Science (CSECS) · doi:10.1109/csecs64665.2025.11009927 · Cited by 2
“Graph Transformers (GTs)[25] improve the transforming power of conventional transformer topologies, which have made substantial progress in natural language processing [26] and computer vision [27].”
Depth-Based Privileged Information for Boosting 3D Human Pose Estimation on RGB
Simoni, A., Marchetti, F., Borghi, G., Becattini, F., Davoli, D., Garattoni, L., Francesca, G., Seidenari, L., & Vezzani, R. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-91575-8_12 · Cited by 4
Li, Y., Ye, Z., Fu, X., et al. (2025) · Proceedings - ACM IEEE Design Automation Conference · doi:10.1109/dac63849.2025.11132426 · Cited by 2
“By partitioning input images into sequences of patches and leveraging self-attention mechanisms, ViTs excel at capturing global image features and have demonstrated superior performance in dense prediction tasks [2], such as object detection, semantic segmentation, and depth estimation.”
Unlocking Comics: The AI4VA Dataset for Visual Understanding
Grönquist, P., Bhattacharjee, D., Aydemir, B., Ozaydin, B., Zhang, T., Salzmann, M., & Susstrunk, S. E. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-92808-6_10 · Cited by 3
Pseudo-trilateral adversarial training for domain adaptive traversability prediction
Chen, Z., Pushp, D., Gregory, J. M., & Liu, L. (2023) · Autonomous Robots · doi:10.1007/s10514-023-10123-7 · Cited by 2
“Recent state-of-the-art works (Ranftl et al., 2021; Zheng et al., 2021; Xie et al., 2021) for semantic segmentation heavily rely on Transformer structure.”
BackgroundWang, L., Lu, R., Sun, F., Zhang, Y., & Yu, J. (2026) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2026.3668665 · Cited by 2
“Therefore, in our experiments, we try to extract the corresponding canny edge information using the hard edge detection method [22], and use the Dense Prediction Transformer [23] method to extract the depth information from the real images.”
Multi-resolution Fine-Tuning of Vision Transformers
Fitzgerald, K., Law, M., Seah, J. C. Y., Tang, J. S. N., & Matuszewski, B. J. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-12053-4_40 · Cited by 2
Classification of Echocardiographic Standard Views Using a Hybrid Attention-based Approach
Ye, Z., Kumar, Y. J., Goh, O. S., Song, F., & Ni, X. (2022) · Intelligent Automation & Soft Computing · doi:10.32604/iasc.2022.023555 · Cited by 2
Mishra, P. K., Taati, B., Ye, B., Newman, K., Mihailidis, A., Iaboni, A., & Khan, S. S. (2026) · IEEE Transactions on Neural Systems and Rehabilitation Engineering · doi:10.1109/tnsre.2026.3656747 · Cited by 3
“The pretrained dense prediction transformer [16] was used to estimate the depth weight Z during both training and testing.”
Yin, C., Luo, J., Ye, Q., & Zhang, X. (2025) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2025.3635418 · Cited by 4
“Second, the self-attention mechanism e ff ectively models spatial relationships and view-specific transformations [67], [70].”
Yıldırım, E., Çölkesen, İ., & Sefercik, U. G. (2026) · Precision Agriculture · doi:10.1007/s11119-026-10362-5 · Cited by 2
“Dense Prediction Transformers (DPT), introduced by Ranftl et al. (2021), are encoder-decoder architectures that employ vision transformers as the primary backbone for dense prediction tasks, including semantic segmentation and monocular depth estimation.”
Transformers in Unsupervised Structure-from-Motion
Chawla, H., Varma, A., Arani, E., & Zonooz, B. (2023) · Communications in computer and information science · doi:10.1007/978-3-031-45725-8_14 · Cited by 3
“Depth Network For the depth network, we use a transformer-based architecture in the encoder, and readapt the decoder from the DPT [46].”
MethodDecTrain: Deciding When to Train a Monocular Depth DNN Online
Fu, Z.-S., Sudhakar, S., Karaman, S., et al. (2025) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2025.3536206 · Cited by 2
“Unlike previous works where either no online training is performed ( N train = 0 ) [8], [9], [10] as seen by the blue curve in Fig.”
Bokeh Effect Rendering with Vision TransformersPreprint
Nagasubramaniam, H., & Younes, R. (2022) · doi:10.36227/techrxiv.17714849.v1 · Cited by 2
Generalizable Thermal-based Depth Estimation via Pre-trained Visual Foundation Model
Fan, R., Wang, Z., Lin, M. G., Wang, Q., Liu, Y., & Wang, W. (2024) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra57147.2024.10610394 · Cited by 2
“To further elaborate on the network architecture, we extend the capabilities of the transformer backbone for thermal-based depth estimation by incorporating an output head inspired by the Dense Prediction Vision Transformer (DPT) [36].”
MethodIPE Transformer for Depth Completion with Input-Aware Positional Embeddings
Li, B., Li, G., Wang, H., et al. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-88013-2_22 · Cited by 2
Zhao, S., Lyu, J., Liu, S., Feng, Z., Ling, H., Jiao, J., Ni, Z., Yang, B., Yao, Q. Q., & Luo, J. (2025) · Frontiers in Plant Science · doi:10.3389/fpls.2025.1687282 · Cited by 3
GeoAvatar: Geometrically-Consistent Multi-Person Avatar Reconstruction from Sparse Multi-View Videos
Lee, S., Kim, S., Lee, H.-K., et al. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.01969 · Cited by 2
“Exploring Monocular Priors Monocular prior models for geometry estimation [1, 5, 8, 11, 19, 35] and segmentation [34] have been proposed by training networks with large image datasets of general scenes.”
Switchable-Encoder-Based Self-Supervised Learning Framework for Monocular Depth and Pose Estimation
Kim, J., Gao, R. Q., Park, J., Yoon, J., & Cho, K. (2023) · Remote Sensing · doi:10.3390/rs15245739 · Cited by 2
RePoseD: Efficient Relative Pose Estimation With Known Depth Information
Ding, Y., Kocur, V., Vávra, V., Haladová, Z. B., Yang, J., Sattler, T., & Kúkelová, Z. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.01380 · Cited by 9
“(2) MDEs typically produce scale/affine-invariant depth [9, 34, 48, 64], affine-invariant inverse depth [8, 54, 67, 68] or metric depth [7, 30, 53], meaning that the estimated depth or inverse depth is only defined up to an unknown scale factor or both unknown scale and shift parameters.”
Adaptive Depth-Converted-Scale Convolution for Self-Supervised Monocular Depth Estimation
Gao, Y., Bai, H., Zhou, H., et al. (2025) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2025.3580746 · Cited by 2
“In this section, we present a brief review of the related studies for deep learning based monocular depth estimation, including the supervised depth estimation methods [22], [23], [24], [25], [26], [27], [28] and self-supervised ones [29], [30], [31], [32], [33], [34], [35], [36].”
SAIL-Recon: Large SfM by Augmenting Scene Regression with Localization
Deng, J., Li, H. L., Xie, T., Ren, W., Zhang, Q., Tan, P., & Guo, X. (2026) · International Conference on 3D Vision (3DV) · doi:10.1109/3dv69130.2026.00037 · Cited by 5
“Inspired by the fact that the network T θ takes a set of 2D images as input and progressively enhances them to compute 3D structures, we extract intermediate feature tokens from each attention layer of T θ to form the scene representation R .”
A depth-estimation-based method for multi-view synthesis applied to Chinese landscape paintings
Peng, X., Zhou, W., Hu, Q., Li, T., Zhang, D., & Cao, R. (2025) · Applied Soft Computing · doi:10.1016/j.asoc.2025.113858 · Cited by 4
Next-Generation Computer Vision in Veterinary Medicine: A Study on Canine Ophthalmology
Burić, M., & Ivašić-Kos, M. (2025) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2025.3530380 · Cited by 3
“Similarly, Dense Prediction Transformer (DPT) [21] and PSPNet [22] use multi-scale context aggregation to enhance segmentation performance across scales.”
Papa, L., Russo, P., & Amerini, I. (2022) · IEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea) · doi:10.1109/metrosea55331.2022.9950812 · Cited by 3
Lopes, C. B. O., Maffei, R. D. Q., & Kolberg, M. (2024) · Journal of Intelligent & Robotic Systems · doi:10.1007/s10846-024-02131-1 · Cited by 2
“The loss of resolution is critical for dense prediction where features should be resolved at the input image resolution level for high performance [26].”
InfluentialLiu, Y., Xue, F., Ming, A., Zhao, M., Ma, H., & Sebe, N. (2024) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3664647.3681405 · Cited by 3
“Early works [29, 39, 40] mainly achieved this goal by training with more accessible relative depth data.”
Radar Fusion Monocular Depth Estimation Based on Dual Attention
Long, J., Huang, J., & Wang, S. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-06794-5_14 · Cited by 2
DAS-COD: Depth-Aware Camouflaged Object Detection via Swin Transformer
Lu, C. Y., Tan, M., Gao, Z. P., et al. (2024) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc54092.2024.10831901 · Cited by 2
“A. Experiment Setup 1) Dataset: We choose the state-of-the-art depth estimation model [21] to offer promising depth at the target.”
Xia, C., Zhang, M., Liu, J., et al. (2025) · Displays · doi:10.1016/j.displa.2025.103104 · Cited by 2
Rethinking Feature Context in Learning Image-Guided Depth Completion
Ding, Y., Li, P., Huang, D., & Li, Z. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-44213-1_9 · Cited by 2
MST: Adaptive Multi-Scale Tokens Guided Interactive Segmentation
Xu, L., Chen, Y., Li, S., & Luo, J. (2025) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2025.3546634 · Cited by 2
“Ranftl et al. [17] upscaled the low-resolution feature maps of ViT to high-resolution for dense prediction tasks.”
BackgroundDiffusionLight-Turbo: Accelerated Light Probes for Free via Single-Pass Chrome Ball Inpainting
Chinchuthakun, W., Phongthawee, P., Raj, A., Jampani, V., Khungurn, P., & Suwajanakorn, S. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3660066 · Cited by 3
“We first predict a depth map from the input image using an off-the-shelf model [57], [58].”
Enhancing monocular height estimation via sparse LiDAR-guided correction
Song, J. H., Chen, H., & Yokoya, N. (2025) · ISPRS Journal of Photogrammetry and Remote Sensing · doi:10.1016/j.isprsjprs.2025.12.004 · Cited by 3
Fu, D., SHANG, X., Hu, Z., et al. (2025) · Journal of Food Science · doi:10.1111/1750-3841.70382 · Cited by 2
Metadata
- Authors
- Ranftl, R., Bochkovskiy, A., & Koltun, V.
- Affiliations
- Intel Labs
- Intel (United States) (US)
- Venue
- 2021 IEEE/CVF International Conference on Computer Vision (ICCV)
- Publisher
- IEEE
- Published
- 2021-10
- Pages
- 12159-12168
- OpenAlex
- W3136635488
- Cited by
- 2,373
- References
- 88
- Citations by year
- 2026: 480 · 2025: 607 · 2024: 586 · 2023: 431 · 2022: 225 · 2021: 44
- FWCI
- 104.94
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE/CVF International Conference on Computer Vision (ICCV) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- TIB Data Manager
- IEEE/CVF International Conference on Computer Vision (ICCV)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Vision and Imaging
- Keywords
- vision transformer, dense prediction, monocular depth estimation, semantic segmentation, convolutional decoder, global receptive field, Pascal Context, NYUv2, KITTI, ADE20K
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
