Research map / Detection and segmentation / Open-set and few-shot detection
High-Quality Entity Segmentation and Grounding
Lu Qi, Yiwen Chen, T B Zhang, Xiangtai Li, Xu Yang, Bo Du, 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 |
|---|---|---|---|
| pretrain Mask2Former R-50 on COCO instead of ImageNet | Mask2Former R-50 · EntitySeg entity segmentation | AP^e: 26.0 -> 35.2 | Improved |
| use a stronger backbone (Swin-T to Swin-L) with COCO pretraining | Mask2Former · EntitySeg entity segmentation | AP^e: 42.9 -> 46.2 | Improved |
| increase single-scale input resolution from (800,1333) to (2700,4500) | Mask2Former · EntitySeg entity segmentation | AP^e: 39.5 -> 40.9 (train time 2.4 -> 60.8 GPU days) | Improved |
| add CropFormer full-image plus crop fusion with association module and batch decoder (ablation) | Mask2Former · EntitySeg entity segmentation | AP^e: 39.5 -> 41.0 | Improved |
| replace CropFormer fusion with video-level Mask2Former or VITA (ablation) | crop fusion · EntitySeg entity segmentation | AP^e: 39.7 and 39.9 vs 41.0 for CropFormer | No clear effect |
| pretrain on EntitySeg instead of COCO-Panoptic then fine-tune on ADE20K | Mask2Former · ADE20K transfer | ENT AP^e: +1.5 | Improved |
| pretrain on EntitySeg instead of COCO-Panoptic then fine-tune on UVO video | Mask2Former · UVO transfer | ENT AP^e: 29.5 (+2.2) | Improved |
| train on EntitySeg instead of COCO entity annotations | Mask2Former Swin-L · In-the-wild avg AR@100 | AR@100: 52.23 -> 67.41 | Improved |
| scale GELLA language model from 7B to 13B | GELLA · RefCOCO val | cIoU: 76.1 -> 76.7 | No clear effect |
| scale GELLA language model from 7B to 13B | GELLA · PNG | AR: 69.8 -> 71.3 | Improved |
| scale GELLA language model from 7B to 13B | GELLA · COCO panoptic segmentation (50 epochs) | PQ: 56.4 -> 56.5 | No clear effect |
| train GELLA 50 epochs instead of 12 | GELLA-7B · COCO panoptic segmentation | PQ: 53.1 -> 56.4 | 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.
- IS-Net — ISNet: Integrate Image-Level and Semantic-Level Context for Semantic Segmentation (ICCV 2021) · 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
- Exploring Simple Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.86)
- Mask Grounding for Referring Image Segmentation (CVPR 2024 · similarity 0.85)
- Training-free Open-Vocabulary Semantic Segmentation via Diverse Prototype Construction and Sub-region Matching (AAAI 2025 · similarity 0.85)
- Open-Vocabulary Semantic Segmentation with Image Embedding Balancing (CVPR 2024 · similarity 0.85)
- EntitySAM: Segment Everything in Video (CVPR 2025 · similarity 0.84)
- Effective SAM Combination for Open-Vocabulary Semantic Segmentation (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.
