Research map / Detection and segmentation / Open-set and few-shot detection
Extract Free Dense Labels from CLIP
Chong Zhou, Chen Change Loy, Bo Dai
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 key smoothing (ablation) | MaskCLIP (RN50) · PASCAL Context | mIoU: 18.5 -> 21.0 | Improved |
| add prompt denoising (ablation) | MaskCLIP (RN50) · PASCAL Context | mIoU: 18.5 -> 19.0 | Improved |
| add key smoothing and prompt denoising (ablation) | MaskCLIP (ViT-B/16) · PASCAL Context | mIoU: 21.7 -> 25.5 | Improved |
| add key smoothing and prompt denoising (ablation) | MaskCLIP (ViT-B/16) · COCO-Stuff | mIoU: 12.5 -> 14.6 | Improved |
| use value features instead of attention in last layer (MaskCLIP vs baseline) (ablation) | CLIP ViT-B/16 · PASCAL Context | mIoU: 9.0 -> 21.7 | Improved |
| train with MaskCLIP pseudo labels (MaskCLIP+) | MaskCLIP (ViT-B/16) · PASCAL Context | mIoU: 25.5 -> 31.1 | Improved |
| use ViT-B/16 instead of ResNet-50 | MaskCLIP · PASCAL Context | mIoU: 18.5 -> 21.7 | Improved |
| add MaskCLIP-guided learning (ablation) | adapted DeepLabv2 · PASCAL VOC zero-shot | mIoU(U): 3.7 -> 72.8 | Improved |
| add self-training (ablation) | MaskCLIP+ · PASCAL VOC zero-shot | mIoU(U): 72.8 -> 86.1 | Improved |
| add self-training (ablation) | MaskCLIP+ · PASCAL VOC zero-shot | mIoU(S): 89.5 -> 88.8 | Hurt |
| multi-scale ensemble of input resolutions (ablation) | MaskCLIP · PASCAL Context | mIoU: 22.72 -> 26.34 | Improved |
| increase input resolution to 520 (ablation) | MaskCLIP · PASCAL Context | mIoU: 22.72 -> 21.68 | No clear effect |
Closest papers by meaning
- Towards Open-Vocabulary Semantic Segmentation Without Semantic Labels (NeurIPS 2024 · similarity 0.91)
- Exploring Regional Clues in CLIP for Zero-Shot Semantic Segmentation (CVPR 2024 · similarity 0.89)
- CLIP as RNN: Segment Countless Visual Concepts without Training Endeavor (CVPR 2024 · similarity 0.89)
- Unveiling the Knowledge of CLIP for Training-Free Open-Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.89)
- Building Vision-Language Models on Solid Foundations with Masked Distillation (CVPR 2024 · similarity 0.89)
- Frozen CLIP: A Strong Backbone for Weakly Supervised Semantic Segmentation (CVPR 2024 · 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.
