CoChat

CGNet: A Light-Weight Context Guided Network for Semantic Segmentation

Open access
Wu, Tianyi, Tang, Sheng, Zhang, Rui, Cao, Juan, Zhang, Yongdong · 2021 · IEEE Transactions on Image Processing

The DOI matches its Crossref record; 66 of the 84 references it cites were found in a registry or an index; 3 of the 23 sources its claims cite were checked for support.

Abstract

The demand of applying semantic segmentation model on mobile devices has been increasing rapidly. Current state-of-the-art networks have enormous amount of parameters hence unsuitable for mobile devices, while other small memory footprint models follow the spirit of classification network and ignore the inherent characteristic of semantic segmentation. To tackle this problem, we propose a novel Context Guided Network (CGNet), which is a light-weight and efficient network for semantic segmentation. We first propose the Context Guided (CG) block, which learns the joint feature of both local feature and surrounding context effectively and efficiently, and further improves the joint feature with the global context. Based on the CG block, we develop CGNet which captures contextual information in all stages of the network. CGNet is specially tailored to exploit the inherent property of semantic segmentation and increase the segmentation accuracy. Moreover, CGNet is elaborately designed to reduce the number of parameters and save memory footprint. Under an equivalent number of parameters, the proposed CGNet significantly outperforms existing light-weight segmentation networks. Extensive experiments on Cityscapes and CamVid datasets verify the effectiveness of the proposed approach. Specifically, without any post-processing and multi-scale testing, the proposed CGNet achieves 64.8% mean IoU on Cityscapes with less than 0.5 M parameters.

Overview

CoChat Summary

TL;DR

CGNet is a lightweight semantic segmentation network built from Context Guided (CG) blocks that combine local features, surrounding context, and global context in all stages. It achieves 64.8% mean IoU on Cityscapes test with less than 0.5 M parameters and no post-processing or multi-scale testing.

Key findings

  • CGNet M3N21 reaches 64.8% mIoU on Cityscapes test with 0.5 M parameters, 6.0 GFLOPs, 334.0 M memory, and 56.8 ms running time at 2048×1024 input.
  • On CamVid test, CGNet M3N21 reaches 65.6% mIoU with 0.5 M parameters, 14.3% higher than ENet (51.3%) and 1.9% lower than FCN (Res101) (67.5%) while using 110× fewer parameters.
  • Ablations on Cityscapes validation: full surrounding context extractor improves mIoU from 54.6% to 59.7%; global context extractor from 58.9% to 59.7%; input injection from 59.4% to 59.7%; PReLU over ReLU from 58.1% to 59.7%; GRL over LRL from 57.2% to 63.5%; removing 1×1 convolution improves from 53.3% to 63.5%.
  • CGNet M3N21 is 131× and 57× smaller than PSPNet and DenseASPP, with category-wise accuracy only 5.4% and 5.5% lower, and memory 10× less than DenseASPP (334.0 M vs. 3997.5 M).

Methods

  • The CG block contains a local feature extractor (3×3 standard convolution), surrounding context extractor (3×3 atrous/dilated convolution), joint feature extractor (concatenation + BN + PReLU), and global context extractor (global average pooling + multilayer perceptron + scale layer), with local or global residual learning.
  • CGNet uses 51 convolutional layers, three downsampling stages with channels 32, 64, and 128, CG blocks in stages 2 and 3, input injection, and a final 1×1 convolution; Cityscapes configuration uses M=3 and N=21.
  • Experiments use Cityscapes (2,975 train / 500 val / 1,525 test images, 19 classes) and CamVid (367 train / 101 val / 233 test images, 11 classes), trained in PyTorch on 2× V100 GPUs with ADAM, poly learning rate, batch size 14, 60K iterations, and single-scale mean IoU evaluation.

Limitations

  • The paper does not state a limitations section.
  • Evident from the text: evaluation is limited to Cityscapes and CamVid, both road-scene datasets, so generalization to other domains is not demonstrated.
  • Evident from the text: ablation studies and architecture choices (M=3, N=21) are tuned on the Cityscapes validation set, and comparisons include some methods using multi-scale testing while CGNet uses no testing tricks.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/tip.2020.3042065

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 Valid · 0 Not valid · 10 Not checked · 43 Own results and interpretations

26 more claims

Create a free account to see all 56 claims and to verify the 18 references still open.

See all claims — free account
  • It may state two separate claims as one. 18
  • The quote only partly states this claim. 6
  • Probably not a substantial claim. 6

Create a free account to see these claims.

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?

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.

2 of 2

Citations

Cites

Research works printed but not linked: 1

  1. liu, 2017, Learning affinity via spatial propagation networks, Proc Adv Neural Inf Process Syst, 1520

Cited by

See all 859

Metadata

Authors
Wu, T., Tang, S., Zhang, R., Cao, J., & Zhang, Y.
Affiliations
Chinese Academy of Sciences (CN)
Venue
IEEE Transactions on Image Processing
Published
2021
Volume
30
Pages
1169-1179
ISSN
  • 1057-7149 (ISSN-L)
  • 1941-0042
OpenAlex
W3112503277
Cited by
859
References
84
Citations by year
2026: 145 · 2025: 222 · 2024: 172 · 2023: 129 · 2022: 97 · 2021: 59
FWCI
30.23
Type
journal article
Language
en
Open access
Locations
  • IEEE Transactions on Image Processing (publishedVersion)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
  • National Key Research and Development Program of China (2017YFC0820605)
  • National Natural Science Foundation of China (61525206, U1703261, 61871004)
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 › Medical Image Segmentation Techniques
Keywords
semantic segmentation, global context, local features, memory footprint reduction, parameter reduction, mobile devices, cityscape, CamVid

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-04, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-04 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

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

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.