Research map / Detection and segmentation / Weakly supervised detection and segmentation
Generative Prompt Model for Weakly Supervised Object Localization
Yuzhong Zhao, Qixiang Ye, Weijia Wu, Chunhua Shen, Fang Wan
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 |
|---|---|---|---|
| learn representative embedding f_r initialized from f_d instead of using f_d only (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 61.2 -> 64.0 | Improved |
| initialize f_r from f_d rather than randomly (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 44.6 -> 64.0 | Improved |
| combine f_r and f_d into f_c (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 64.0 -> 64.5 | Improved |
| finetune the attention-unet backbone on the WSOL dataset (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 64.5 -> 65.1 | Improved |
| aggregate attention maps at timesteps 1 and 100 instead of single timestep (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 64.9 -> 65.1 | Improved |
| add large timestep 500 to the aggregation (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 65.1 -> 64.1 | Hurt |
| aggregate resolutions 8 and 16 instead of 16 alone (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 62.5 -> 65.1 | Improved |
| add 32x32 resolution to the aggregation (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 65.1 -> 64.2 | Hurt |
| scale up discriminative TS-CAM with ViT-H backbone and LAION-2B data | TS-CAM · ImageNet-1K val | Top-1 Loc: 53.4 -> 42.1 | Hurt |
| add noise at inference time (averaged over 10 seeds) (ablation) | GenPromp · ImageNet-1K val | Top-1 Loc: 64.8 -> 65.1 | Improved |
| prompt ensemble over 7 templates | GenPromp · ImageNet-1K val | Top-1 Loc: 65.1 -> 65.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- C2AM — C2AM: Contrastive Learning of Class-Agnostic Activation Map for Weakly Supervised Object Localization and Semantic Segmentation (CVPR 2022) · better on 2 of 2 shared results
- TS-CAM — TS-CAM: Token Semantic Coupled Attention Map for Weakly Supervised Object Localization (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
- Evaluating Weakly Supervised Object Localization Methods Right (CVPR 2020 · similarity 0.90)
- Evaluation for Weakly Supervised Object Localization: Protocol, Metrics, and Datasets (TPAMI 2022 · similarity 0.89)
- Pro2SAM: Mask Prompt to SAM with Grid Points for Weakly Supervised Object Localization (ECCV 2024 · similarity 0.88)
- Pairwise Similarity Knowledge Transfer for Weakly Supervised Object Localization (ECCV 2020 · similarity 0.88)
- Weakly Supervised Object Localization via Transformer with Implicit Spatial Calibration (ECCV 2022 · similarity 0.88)
- Generalizable Object Keypoint Localization from Generative Priors (CVPR 2025 · similarity 0.87)
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.
