Research map / Detection and segmentation / CNN features
BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation
Changqian Yu, Changxin Gao, Jingbo Wang, Gang Yu, Chunhua Shen, Nong Sang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Semantic Branch to Detail Branch (sum) (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 62.35 -> 68.60 | Improved |
| add Detail Branch to Semantic Branch (sum) (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 64.68 -> 68.60 | Improved |
| use Bilateral Guided Aggregation instead of summation (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 68.60 -> 69.67 | Improved |
| use Bilateral Guided Aggregation instead of concatenation (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 68.93 -> 69.67 | Improved |
| add booster training strategy (auxiliary heads) (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 69.67 -> 73.19 | Improved |
| add OHEM online bootstrapping (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 73.19 -> 73.36 | Improved |
| add a very light Semantic Branch (lambda=1/16) (ablation) | Detail Branch · Cityscapes val | mIoU: 62.35 -> 68.27 | Improved |
| use 3x3 Gather Layer instead of 1x1 convolution (ablation) | GE Layer · Cityscapes val | mIoU: 66.62 -> 69.67 | Improved |
| use double depth-wise convs instead of one 5x5 at stride 2 (ablation) | GE Layer · Cityscapes val | mIoU: 68.98 -> 69.67 | Improved |
| add context embedding to the GE layer (ablation) | GE Layer · Cityscapes val | mIoU: 69.01 -> 69.67 | Improved |
| increase expansion ratio from 6 to 8 (ablation) | Semantic Branch · Cityscapes val | mIoU: 69.67 -> 68.99 | Hurt |
| widen the model to alpha=2.0 (ablation) | BiSeNetV2 · Cityscapes val | mIoU: 73.36 -> 75.11 | Improved |
Later papers that compared with this method
3 later accepted papers report BiSeNetV2 as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Hilbert Curve-Based Attention Enabling Topology-Preserving Image Tensor Representation for Semantic Segmentation Network (CVPR 2026 · better on 1 of 1 shared result)
- Not All Pixels are Equal: Learning Pixel Hardness for Semantic Segmentation (IJCV 2025 · better on 1 of 1 shared result)
- Lightweight and Progressively-Scalable Networks for Semantic Segmentation (IJCV 2023 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- ICNet — ICNet: Intra-saliency Correlation Network for Co-Saliency Detection (NeurIPS 2020) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation (ECCV 2018 · similarity 0.96)
- Rethinking BiSeNet for Real-Time Semantic Segmentation (CVPR 2021 · similarity 0.92)
- PIDNet: A Real-Time Semantic Segmentation Network Inspired by PID Controllers (CVPR 2023 · similarity 0.90)
- DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation (CVPR 2019 · similarity 0.89)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 2024 · similarity 0.88)
- BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
