Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
ReCo: Retrieve and Co-segment for Zero-shot Transfer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DenseCLIP refinement at inference (ablation) | ReCo · PASCAL-Context val | mIoU: 5.7 -> 21.8 | Helped |
| add language-guided co-segmentation (ablation) | ReCo + DenseCLIP · PASCAL-Context val | mIoU: 21.8 -> 23.1 | Helped |
| add context elimination (ablation) | ReCo + DenseCLIP · PASCAL-Context val | mIoU: 21.8 -> 26.0 | Helped |
| add CRF post-processing (ablation) | ReCo + DenseCLIP + LGC + CE · PASCAL-Context val | mIoU: 26.6 -> 27.2 | Helped |
| suppress all five context categories instead of none (ablation) | ReCo context elimination · PASCAL-Context val | mIoU: 5.7 -> 12.3 | Helped |
| suppress building after tree and sky (ablation) | ReCo context elimination · PASCAL-Context val | mIoU: 12.0 -> 10.8 | Hurt |
| rephrase parking to parking lot and vegetation to tree (ablation) | ReCo · Cityscapes val | mIoU: 15.4 -> 19.3 | Helped |
| ReCo zero-shot vs DenseCLIP annotation-free baseline | ReCo · COCO-Stuff val | mIoU: 19.6 -> 26.3 | Helped |
| train DeepLabv3+ on ReCo pseudo-labels (ReCo+) vs STEGO | ReCo+ · COCO-Stuff val | pixel accuracy: 56.9 -> 54.1 | Hurt |
| fine-tune on ReCo pseudo-labels (ReCo+ vs ReCo) | ReCo · Cityscapes val | mIoU: 19.3 -> 24.2 | Helped |
Closest papers by meaning
- CLIP as RNN: Segment Countless Visual Concepts without Training Endeavor (CVPR 2024 · similarity 0.90)
- Cascade-CLIP: Cascaded Vision-Language Embeddings Alignment for Zero-Shot Semantic Segmentation (ICML 2024 · similarity 0.89)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.88)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.88)
- Early Semantic Grounding in Image Editing Models for Zero-Shot Referring Image Segmentation (NeurIPS 2026 · similarity 0.88)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.88)
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.