Research map / Detection and segmentation / CNN features
LevelSet R-CNN: A Deep Variational Method for Instance Segmentation
Namdar Homayounfar Yuwen Xiong Justin Liang Wei-Chiu Ma Raquel Urtasun {namdar, yuwen, justin.liang, weichiu, urtasun}@uber.com
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 the Chan-Vese level set module and extra heads | Mask R-CNN · Cityscapes val | AP: 32.3 -> 36.2 | Improved |
| add the level set module | Mask R-CNN · COCO val | AP: 33.8 -> 34.3 | Improved |
| add the level set module | Mask R-CNN · LVIS val (COCO categories) | AP*: 35.6 -> 36.4 | Improved |
| run unrolled level set optimization on top of the initial mask (ablation) | LevelSet R-CNN · Cityscapes val | AP: 35.4 -> 36.2 | Improved |
| run unrolled level set optimization on top of the initial mask (ablation) | LevelSet R-CNN · COCO val | AP: 33.7 -> 34.3 | Improved |
| backpropagate the final TSDF loss through the initial TSDF head (ablation) | LevelSet R-CNN · Cityscapes val | final mask AP: 34.2 -> 36.2 | Improved |
| backpropagate the final TSDF loss through the initial TSDF head (ablation) | LevelSet R-CNN · Cityscapes val | initial mask AP: 33.0 -> 35.4 | Improved |
| add instance-adaptive hyperparameter head instead of global hyperparameters (ablation) | LevelSet R-CNN · Cityscapes val | AP: 35.4 -> 36.2 | Improved |
| use region-based Chan-Vese module instead of level set loss | Mask R-CNN · Cityscapes val | AP: 34.3 -> 36.2 | Improved |
| use region-based Chan-Vese module instead of DELSE edge-based level set | Mask R-CNN · Cityscapes val | AP: 35 -> 36.2 | Improved |
| train mask head with TSDF loss only, without Chan-Vese components (ablation) | Mask R-CNN · Cityscapes val | AP: 32.3 -> 32.3 | No clear effect |
| increase Mask R-CNN mask head resolution to 112x112 | Mask R-CNN · Cityscapes val | AP: drops by 2.2 AP from 32.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 4 on most shared results.
- PANet — PANet: Few-Shot Image Semantic Segmentation With Prototype Alignment (ICCV 2019) · better on 2 of 2 shared results
- Polytransform — PolyTransform: Deep Polygon Transformer for Instance Segmentation (CVPR 2020) · better on 0 of 1 shared result
- SSAP — SSAP: Single-Shot Instance Segmentation With Affinity Pyramid (ICCV 2019) · better on 1 of 1 shared result
- AdaptIS — AdaptIS: Adaptive Instance Selection Network (ICCV 2019) · better on 1 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
- Deep Variational Instance Segmentation (NeurIPS 2020 · similarity 0.89)
- Mask R-CNN (ICCV 2017 · similarity 0.88)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.87)
- RDSNet: A New Deep Architecture forReciprocal Object Detection and Instance Segmentation (AAAI 2020 · similarity 0.87)
- Panoptic Feature Pyramid Networks (CVPR 2019 · similarity 0.87)
- Hybrid Task Cascade for Instance Segmentation (CVPR 2019 · 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.
