Research map / Detection and segmentation / Open-set and few-shot detection
High Quality Entity Segmentation
Lu Qi, Jason Kuen, Tiancheng Shen, Jiuxiang Gu, Wenbo Li, Weidong Guo, Jiaya Jia, Zhe Lin, Ming-Hsuan Yang
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 |
|---|---|---|---|
| fuse full image and high-res crops with CropFormer | Mask2Former · EntitySeg entity segmentation | AP^e: 39.5 -> 41.0 | Improved |
| increase Mask2Former input resolution to (2700,4500) | Mask2Former · EntitySeg entity segmentation | AP^e: 39.5 -> 40.6 | Mixed |
| add CropFormer on longer training schedules | Mask2Former · EntitySeg entity segmentation | AP^e: 40.9 -> 42.8 | Improved |
| use both full image and crop predictions (B-OC) instead of crops only (B-C) (ablation) | CropFormer decoder · EntitySeg entity segmentation | AP^e: 40.2 -> 41.0 | Improved |
| use batch decoder crops (B-C) instead of full image (I-O) (ablation) | CropFormer decoder · EntitySeg entity segmentation | AP^e: 39.3 -> 40.2 | Improved |
| use crop ratio 0.7 instead of 0.5 (ablation) | CropFormer · EntitySeg entity segmentation | AP^e: 38.5 -> 41.0 | Improved |
| use crop ratio 0.8 instead of 0.7 (ablation) | CropFormer · EntitySeg entity segmentation | AP^e: 41.0 -> 40.9 | No clear effect |
| use fixed 4 corner crops instead of random crops in training (ablation) | CropFormer · EntitySeg entity segmentation | AP^e: 39.7 -> 41.0 | Improved |
| test with 8 fixed crops instead of 4 (trained with 4) (ablation) | CropFormer · EntitySeg entity segmentation | AP^e: 41.0 -> 41.3 | Improved |
| add self-attention and FFN to association module cross-attention (ablation) | CropFormer · EntitySeg entity segmentation | AP^e: 40.7 -> 41.0 | Improved |
| add CropFormer to Mask2Former on COCO instance segmentation | Mask2Former · COCO instance segmentation | AP: 50.1 -> 51.9 | Improved |
| add CropFormer to Mask2Former on ADE20K panoptic segmentation | Mask2Former · ADE20K panoptic | PQ: 48.1 -> 49.7 | 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.
- PatchDCT — PatchDCT: Patch Refinement for High Quality Instance Segmentation (ICLR 2023) · 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
- Open World Entity Segmentation (TPAMI 2022 · similarity 0.87)
- LQMFormer: Language-aware Query Mask Transformer for Referring Image Segmentation (CVPR 2024 · similarity 0.86)
- MaSS13K: A Matting-level Semantic Segmentation Benchmark (CVPR 2025 · similarity 0.85)
- PEM: Prototype-based Efficient MaskFormer for Image Segmentation (CVPR 2024 · similarity 0.85)
- Scaling up Image Segmentation across Data and Tasks (CVPR 2025 · similarity 0.85)
- EntitySAM: Segment Everything in Video (CVPR 2025 · similarity 0.84)
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.
