FIVR: Fine-Grained Incident Video Retrieval
Open accessThe DOI matches its Crossref record; 55 of the 59 references it cites were found in a registry or an index; none of the 10 sources its claims cite is checked for support yet.
Abstract
This paper introduces the problem of Fine-grained Incident Video Retrieval (FIVR). Given a query video, the objective is to retrieve all associated videos, considering several types of associations that range from duplicate videos to videos from the same incident. FIVR offers a single framework that contains several retrieval tasks as special cases. To address the benchmarking needs of all such tasks, we construct and present a large-scale annotated video dataset, which we call FIVR-200K, and it comprises 225,960 videos. To create the dataset, we devise a process for the collection of YouTube videos based on major news events from recent years crawled from Wikipedia and deploy a retrieval pipeline for the automatic selection of query videos based on their estimated suitability as benchmarks. We also devise a protocol for the annotation of the dataset with respect to the four types of video associations defined by FIVR. Finally, we report the results of an experimental study on the dataset comparing five state-of-the-art methods developed based on a variety of visual descriptors, highlighting the challenges of the current problem.
Overview
CoChat Summary
TL;DR
This paper introduces Fine-grained Incident Video Retrieval (FIVR), a unified retrieval problem covering duplicate, complementary-viewpoint, and same-incident video associations. It presents FIVR-200K, a 225,960-video benchmark built from Wikipedia news events, and benchmarks five state-of-the-art retrieval methods across three FIVR tasks.
Key findings
- FIVR-200K contains 225,960 YouTube videos, 4,687 Wikipedia news events, 100 queries, and 7,100 hours of video; selected queries average 13 NDV, 57 DSV, 18 CSV, and 35 ISV annotations.
- The best benchmarked method, LBoW, achieves mAP scores of 0.710 for DSVR, 0.675 for CSVR, and 0.572 for ISVR; GV performs worst at 0.165, 0.153, and 0.118, respectively.
- All methods perform much better on CC WEB VIDEO, with mAP values as high as 0.976, showing FIVR-200K is substantially more challenging, especially for ISVR.
- Deep features outperform handcrafted features; LBoW with VGG features is best for DSVR/CSVR, while ISVR remains the hardest task with most queries below 0.6 mAP.
Methods
- Dataset construction: crawled Wikipedia “Current Events” from Jan 1, 2013 to Dec 31, 2017, retained “Armed conflicts and attacks” or “Disasters and accidents” categories (9,431 collected, 4,687 retained), then used YouTube API with event headlines, filtering to videos published from event start up to one week after and duration up to five minutes.
- Query selection: computed visual and textual similarity via tf-idf BoW representations, built a video graph with threshold 0.7, discarded components with uploader ratio below 0.75, retained components within a two-week publication window, selected videos under 90 seconds, chose earliest-published video per component, and kept the top 100 queries by component size.
- Annotation and evaluation: annotated videos in three phases using labels ND, DS, CS, IS, and DI; benchmarked GV, BoW, LBoW, DML, and HC with handcrafted, CNN, and 3D CNN features; evaluated DSVR, CSVR, and ISVR using interpolated PR curves and mAP.
Limitations
- Stated: global video representations have strong performance limits; suggested improvements such as frame-level matching, end-to-end CNN/3D CNN training, and query expansion are computationally complex and need careful design for large-scale use.
- Stated: the dataset annotation covers four labels, and the authors plan to extend it for related problems such as event reconstruction and synchronization.
- Evident from the text: annotation was split between two annotators with only the lead author revisiting non-distractor labels for consistency; the query set is limited to 100 selected from 635 candidates, and the dataset is restricted to two Wikipedia event categories, YouTube videos up to five minutes, and events from 2013–2017.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 0 Not valid · 6 Not checked · 98 Own results and interpretations
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“To accurately represent the problem, this dataset is composed of user-generated videos related to a large number of real-world events.”
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.
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
“FIVR-200K | 100 | 225,960 | 7,100 | ✓ | Fine-grained Incident Video Retrieval”
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
“however, the performance gap with deep features is considerable.”
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
“In particular, LBoW achieves a mAP score of 0.710 in the DSVR task”
- Also stated in: § B. Experiments · p. 12 · plain
“the LBoW aggregation scheme in combination with the VGG CNN features achieves the best result (mAP = 0 : 710 )”
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.
- Also stated in: § B. Experiments · p. 12 · plain
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
“and 3KB per video to store the video representations.”
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.
54 more claims
Create a free account to see all 104 claims and every source verdict.
- Probably not a substantial claim. 30
- It may state two separate claims as one. 12
- The quote only partly states this claim. 3
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
§ B. Experiments · p. 13
“The results of all methods have been measured using the open source library Scikit-learn [47] in Python, on a Linux PC with a 4-core i7-4770K and 32GB of RAM.”
§ REFERENCES · p. 15
“[47] S. Van der Walt, J. L. Sch¨ onberger, J. Nunez-Iglesias, F. Boulogne, J. D. Warner, N. Yager, E. Gouillart, and T. Yu, 'scikit-image: image processing in python,' PeerJ , vol. 2, p. e453, 2014.”
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
§ A. Video Datasets · p. 3
“A more recent dataset that is relevant to our problem is VCDB [20].”
§ REFERENCES · p. 14
“[20] Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell, 'Caffe: Convolutional architecture for fast feature embedding,' in Proceedings of the 22nd ACM international conference on Multimedia . ACM, 2014, pp. 675-678.”
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 2×
“• Intermediate CNN Features: We employ three popular architectures for the extraction of intermediate CNN features [27]: VGG-16 (VGG) [40], ResNet-152 (RES) [17] and Inception-V4 (INC) [44].”
MethodVery Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
Distinctive Image Features from Scale-Invariant Keypoints
Lowe, D. G. (2004) · International Journal of Computer Vision · doi:10.1023/b:visi.0000029664.99615.94 · Cited by 55,589 · cited 1×
“Early approaches employed handcrafted features including HSV Colour Histograms [16], [25], [42], [49], Local Binary Patterns (LBP) [25], [37], [42], [50], [51] fuzzy multidimensional histograms of colour and motion video segments [10], Auto Colour Correlograms (ACC) [7], [18], and several keypoint descriptors (such as…”
MethodSURF: Speeded Up Robust Features
Bay, H., Tuytelaars, T., & Van Gool, L. (2006) · Lecture Notes in Computer Science · doi:10.1007/11744023_32 · Cited by 14,955
Caffe: Convolutional Architecture for Fast Feature Embedding
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., & Darrell, T. (2014) · Proceedings of the 22nd ACM international conference on Multimedia · doi:10.1145/2647868.2654889 · Cited by 11,077 · cited 8×
“More specifically, related datasets include CC WEB VIDEO [49], UQ VIDEO [42], MUSCLEVCD [30], TRECVID-CBCD 2011 [29], VCDB [21] and EVVE [35].”
BackgroundInfluentialQuo Vadis, Action Recognition? A New Model and the Kinetics Dataset
Carreira, J., & Zisserman, A. (2017) · Computer Vision and Pattern Recognition · doi:10.48550/arxiv.1705.07750 · Cited by 10,485
Learning Spatiotemporal Features with 3D Convolutional Networks
Tran, D., Bourdev, L., Fergus, R., Torresani, L., & Paluri, M. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.510 · Cited by 9,593 · cited 2×
“3D CNN Features: We employ two popular architectures for the extraction of 3D CNN features: C3D [46] and I3D [8].”
MethodQuo Vadis, Action Recognition? A New Model and the Kinetics Dataset
Carreira, J., & Zisserman, A. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.502 · Cited by 9,570 · cited 6×
“• 3D CNN Features: We employ two popular architectures for the extraction of 3D CNN features: C3D [46] and I3D [8].”
MethodInfluentialscikit-image: image processing in Python
van der Walt, S., Schönberger, J. L., Nunez-Iglesias, J., Boulogne, F., Warner, J. D., Yager, N., Gouillart, E., & Yu, T. (2014) · PeerJ · doi:10.7717/peerj.453 · Cited by 6,794
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
Szegedy, C., Ioffe, S., Vanhoucke, V., & Alemi, A. (2017) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v31i1.11231 · Cited by 4,636
Research works printed but not linked: 2
- silverman, 2013, Verification handbook
- kraaij, 0, TRECVID 2011 content-based copy detection: Task overview, Online Proc TRECVid 2010
Cited by
ViSiL: Fine-Grained Spatio-Temporal Video Similarity Learning
Kordopatis-Zilos, G., Papadopoulos, S., Patras, I., & Kompatsiaris, I. Y. (2019) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00645 · Cited by 109
Self-citationTemporal Context Aggregation for Video Retrieval with Contrastive Learning
Shao, J., Wen, X., Zhao, B., & Xue, X. (2021) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv48630.2021.00331 · Cited by 91
“Extensive experiments are conducted on multi video retrieval datasets, such as CC WEB VIDEO [61], FIVR [30], and EVVE [45].”
MethodInfluentialA hierarchical network-oriented analysis of user participation in misinformation spread on WhatsApp
Nobre, G. P., Ferreira, C. H. G., & Almeida, J. M. (2021) · Information Processing & Management · doi:10.1016/j.ipm.2021.102757 · Cited by 81
“As future work, we intend to explore more sophisticated heuristics, such as word [43] and sentence embeddings [44] (for textual messages), product quantization [45] (for images) and techniques based on audio and video content analysis [46].”
BackgroundCooper, N., Bernal-Cárdenas, C., Chaparro, O., Moran, K., & Poshyvanyk, D. (2021) · ACM/IEEE International Conference on Software Engineering: Companion Proceedings (ICSE-Companion) · doi:10.1109/icse43902.2021.00091 · Cited by 81
DnS: Distill-and-Select for Efficient and Accurate Video Indexing and Retrieval
Kordopatis-Zilos, G., Tzelepis, C., Papadopoulos, S., Kompatsiaris, I. Y., & Patras, I. (2022) · International Journal of Computer Vision · doi:10.1007/s11263-022-01651-3 · Cited by 52
Self-citation
Metadata
- Authors
- Kordopatis-Zilos, G., Papadopoulos, S., Patras, I., & Kompatsiaris, I.
- Affiliations
- Information Technologies Institute (GR)
- Queen Mary University of London (GB)
- Venue
- IEEE Transactions on Multimedia
- Published
- 2019-10
- Volume
- 21
- Issue
- 10
- Pages
- 2638-2652
- ISSN
- 1520-9210 (ISSN-L)
- 1941-0077
- OpenAlex
- W2949298458
- Cited by
- 79
- References
- 59
- Citations by year
- 2026: 12 · 2025: 6 · 2024: 12 · 2023: 13 · 2022: 10 · 2021: 15 · 2020: 5 · 2019: 6
- FWCI
- 3.57
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Transactions on Multimedia (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Queen Mary Research Online (Queen Mary University of London) (submittedVersion)
- Zenodo (CERN European Organization for Nuclear Research) (submittedVersion, cc-by)
- arXiv (Cornell University)
- Funders
- InVID and WeVerify Projects
- European Commission (687786, 825297)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Video Analysis and Summarization
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Retrieval and Classification Techniques
- Keywords
- benchmark dataset, news events, video annotation, image descriptors
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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: FIVR: Fine-Grained Incident Video Retrieval. https://app.cochat.ai/papers/10.1109/tmm.2019.2905741
- The paper
- Kordopatis-Zilos, G., Papadopoulos, S., Patras, I., & Kompatsiaris, I. (2019). FIVR: Fine-Grained Incident Video Retrieval. IEEE Transactions on Multimedia. https://doi.org/10.1109/tmm.2019.2905741
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.