Research map / Detection and segmentation / Open-set and few-shot detection
HIDA: A Human-Intuition-Guided Depth-Aware Framework for Zero-Shot Amodal Segmentation
Peng Li, Kelin Wang, Bingchuan Chen, Ya-Li Hou, Mingxia Shen, Bo Li
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 trainable ViT encoder features to SAM encoder features (ablation) | HIDA (zero-shot) · COCOA-cls val | mIoU (%): 60.90 -> 83.16 | Improved |
| add SAM encoder features to ViT features (ablation) | HIDA (zero-shot) · COCOA-cls val | mIoU (%): 81.56 -> 83.16 | Improved |
| add pseudo-depth map input (ablation) | HIDA (zero-shot) · COCOA-cls val | mIoU (%): 81.78 -> 83.16 | Improved |
| increase soft MoE experts from 1 to 4 (ablation) | HIDA soft MoE · COCOA-cls val | mIoU (%): 81.98 -> 83.16 | Improved |
| increase experts from 4 to 8 (ablation) | HIDA soft MoE · COCOA-cls val | mIoU (%): 83.16 -> 82.97 | No clear effect |
| increase ViT encoder depth from 4 to 8 layers (ablation) | HIDA · COCOA-cls val | mIoU (%): 82.18 -> 83.16 | Improved |
| increase ViT encoder depth from 6 to 8 layers (ablation) | HIDA · COCOA-cls val | mIoU (%): 83.05 -> 83.16 | No clear effect |
| zero-shot HIDA vs SAMBA across datasets | HIDA (zero-shot) · COCOA-cls, D2SA, KINS | mIoU (%): COCOA-cls 83.16 vs 81.82; D2SA 91.11 vs 90.98; KINS 86.71 vs 88.47 | Mixed |
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.
- Pix2Gestalt — pix2gestalt: Amodal Segmentation by Synthesizing Wholes (CVPR 2024) · 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
- Towards Efficient Foundation Model for Zero-shot Amodal Segmentation (CVPR 2025 · similarity 0.88)
- Segment Anything, Even Occluded (CVPR 2025 · similarity 0.87)
- Amodal Ground Truth and Completion in the Wild (CVPR 2024 · similarity 0.87)
- pix2gestalt: Amodal Segmentation by Synthesizing Wholes (CVPR 2024 · similarity 0.86)
- BLADE: Box-Level Supervised Amodal Segmentation through Directed Expansion (AAAI 2024 · similarity 0.86)
- Using Diffusion Priors for Video Amodal Segmentation (CVPR 2025 · similarity 0.85)
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.
