Research map / Detection and segmentation / CNN features
SegStereo: Exploiting Semantic Information for Disparity Estimation
Guorun Yang, Hengshuang Zhao, Jianping Shi, Zhidong Deng, Jiaya Jia
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 feature embedding (PSPNet features) to ResNetCorr (ablation) | photometric-loss stereo net · KITTI Stereo 2015 (unsupervised, all pixels) | EPE: 3.36 -> 2.72 | Improved |
| add semantic feature embedding (ablation) | photometric-loss stereo net · KITTI Stereo 2015 (unsupervised, all pixels) | D1 error: 14.08 -> 12.08 | Improved |
| add semantic feature embedding (ablation) | photometric + smooth loss net · KITTI Stereo 2015 (unsupervised, all pixels) | EPE: 2.43 -> 2.17 | Improved |
| add smoothness loss (ablation) | SegStereo (photometric) · KITTI Stereo 2015 (unsupervised, all pixels) | EPE: 2.72 -> 2.17 | Improved |
| add semantic softmax loss (ablation) | SegStereo (photometric + smooth) · KITTI Stereo 2015 (unsupervised, all pixels) | EPE: 2.17 -> 1.89 | Improved |
| add semantic softmax loss (ablation) | SegStereo (photometric + smooth) · KITTI Stereo 2015 (unsupervised, all pixels) | D1 error: 10.53 -> 10.03 | Improved |
| fine-tune on 200 KITTI 2015 training images | SegStereo unsupervised · KITTI Stereo 2015 (unsupervised, all pixels) | D1 error: 10.03 -> 8.79 | Improved |
| add semantic feature embedding (pretrain on Cityscapes) (ablation) | supervised stereo net · KITTI 2015 test split | D1 error (All): 7.64 -> 7.34 | Improved |
| add semantic feature embedding (pretrain on extra Cityscapes + FlyingThings3D) (ablation) | supervised stereo net · KITTI 2015 test split | D1 error (All): 5.64 -> 5.38 | Improved |
| add semantic feature embedding after finetuning (ablation) | supervised stereo net · KITTI 2012+2015 finetuned, validation split | EPE (All): 0.76 -> 0.75 | No clear effect |
| use finer shallow layer conv1_3 with correlation 96 (corr13) | SegStereo · KITTI 2012+2015 finetuned, validation split | D1 error (All): 2.30 -> 2.25 | Improved |
| add semantic cues to ResNetCorr (ablation) | supervised stereo net · FlyingThings3D test | EPE: 3.50 -> 1.45 | Improved |
Later papers that compared with this method
1 later accepted paper reports SegStereo as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Semantic Stereo Matching With Pyramid Cost Volumes (ICCV 2019 · better on 3 of 3 shared results)
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
- DispViT: Direct Stereo Disparity Regression with a Single-Stream Vision Transformer (ICLR 2026 · similarity 0.88)
- SemStereo: Semantic-Constrained Stereo Matching Network for Remote Sensing (AAAI 2025 · similarity 0.88)
- EdgeStereo: An Effective Multi-task Learning Network for Stereo Matching and Edge Detection (IJCV 2020 · similarity 0.87)
- Disp R-CNN: Stereo 3D Object Detection via Shape Prior Guided Instance Disparity Estimation (CVPR 2020 · similarity 0.87)
- Shape Prior Guided Instance Disparity Estimation for 3D Object Detection (TPAMI 2021 · similarity 0.87)
- SENSE: A Shared Encoder Network for Scene-Flow Estimation (ICCV 2019 · similarity 0.86)
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.
