Research map / Detection and segmentation / Weakly supervised detection and segmentation
Learning Multi-Modal Class-Specific Tokens for Weakly Supervised Dense Object Localization
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 multi-modal class-specific tokens and correlate them with patches (ablation) | Multi-class token transformer · PASCAL VOC 2012 train | mIoU: 62.7 -> 64.1 | Helped |
| Use GWRP instead of GAP pooling (ablation) | Class-to-patch correlation · PASCAL VOC 2012 train | mIoU: 62.1 -> 62.7 | Helped |
| Fuse global and local multi-modal class-specific tokens (ablation) | Multi-modal correlation localization · PASCAL VOC 2012 train | mIoU: 64.1 -> 66.3 | Helped |
| Use local/output rather than global/input multi-modal tokens (ablation) | Multi-modal correlation localization · PASCAL VOC 2012 train | mIoU: 64.1 -> 63.3 | Hurt |
| Add sample-specific visual context (ablation) | Multi-modal class-specific tokens · PASCAL VOC 2012 train | mIoU: 64.1 -> 64.8 | Helped |
| Use CLIP caption embeddings with L1 regularization (ablation) | Visual-context-enhanced token learning · PASCAL VOC 2012 train | mIoU: 64.8 -> 63.7 | Hurt |
| Use batch-contrast cross-entropy instead of L1 for caption embeddings (ablation) | Visual-context-enhanced token learning · PASCAL VOC 2012 train | mIoU: 63.7 -> 65.1 | Helped |
| Regularize text tokens against CLIP image embeddings using batch-contrast CE (ablation) | Visual-context-enhanced token learning · PASCAL VOC 2012 train | mIoU: 65.1 -> 66.3 | Helped |
| Replace the single fixed prompt with prompt ensembling (ablation) | Text-token initialization · PASCAL VOC 2012 train | mIoU: 66.3 -> 65.5 | Hurt |
| Replace the single fixed prompt with learned CoOp prompts (ablation) | Text-token initialization · PASCAL VOC 2012 train | mIoU: 66.3 -> 66.0 | Hurt |
| Use the proposed multi-modal class-specific token method | WSDOL · PASCAL VOC 2012 train | mIoU: 62.3 -> 66.3 | Helped |
| Use the proposed multi-modal class-specific token method | WSDOL · MS COCO 2014 train | mIoU: 37.2 -> 40.9 | Helped |
Closest papers by meaning
- Multi-Class Token Transformer for Weakly Supervised Semantic Segmentation (CVPR 2022 · similarity 0.93)
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation (CVPR 2018 · similarity 0.89)
- MoRe: Class Patch Attention Needs Regularization for Weakly Supervised Semantic Segmentation (AAAI 2025 · similarity 0.89)
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.89)
- Leveraging Class Distributions in CLIP for Weakly Supervised Semantic Segmentation (CVPR 2026 · similarity 0.89)
- Class Tokens Infusion for Weakly Supervised Semantic Segmentation (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.