Research map / Detection and segmentation / CNN features
Instance and Panoptic Segmentation Using Conditional Convolutions
Zhi Tian, Bowen Zhang, Hao Chen, Chunhua Shen
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 relative coordinates to the mask head input (ablation) | CondInst mask head · COCO val2017 | mask AP: 31.5 -> 35.6 | Improved |
| use absolute instead of relative coordinates (ablation) | CondInst mask head · COCO val2017 | mask AP: 35.6 -> 32.0 | Hurt |
| add bottom-branch features to mask head input (ablation) | CondInst mask head · COCO val2017 | mask AP: 31.3 -> 35.6 | Improved |
| increase mask head depth from 1 to 3 (ablation) | CondInst mask head · COCO val2017 | mask AP: 30.5 -> 35.6 | Improved |
| increase mask head depth from 3 to 4 (ablation) | CondInst mask head · COCO val2017 | mask AP: 35.6 -> 35.6 | No clear effect |
| increase mask head width from 2 to 8 (ablation) | CondInst mask head · COCO val2017 | mask AP: 33.9 -> 35.6 | Improved |
| increase mask head width from 8 to 16 (ablation) | CondInst mask head · COCO val2017 | mask AP: 35.6 -> 35.7 | No clear effect |
| vary bottom channels C_bottom from 2 to 16 (ablation) | CondInst bottom branch · COCO val2017 | mask AP: 34.9 -> 35.6 -> 35.4 | No clear effect |
| use P2 instead of P3 for bottom module (ablation) | CondInst · COCO val2017 | mask AP: 35.6 -> 36.0 | Improved |
| add upsampling of mask predictions (factor 2) (ablation) | CondInst · COCO val2017 | mask AP: 34.6 -> 35.6 | Improved |
| use mask-based NMS instead of box NMS (ablation) | CondInst · COCO val2017 | mask AP: 35.6 -> 35.6 | No clear effect |
| use panoptic-generated instance targets instead of instance supervision (ablation) | CondInst panoptic · COCO panoptic | PQ: 42.94 -> 44.18 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- YOLACT-700 — YOLACT: Real-Time Instance Segmentation (ICCV 2019) · better on 2 of 2 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
- Conditional Convolutions for Instance Segmentation (ECCV 2020 · similarity 0.97)
- Fully Convolutional Instance-Aware Semantic Segmentation (CVPR 2017 · similarity 0.90)
- BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation (CVPR 2020 · similarity 0.90)
- PointINS: Point-Based Instance Segmentation (TPAMI 2021 · similarity 0.90)
- K-Net: Towards Unified Image Segmentation (NeurIPS 2021 · similarity 0.90)
- SOLO: Segmenting Objects by Locations (ECCV 2020 · similarity 0.89)
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.
