Research map / Label-efficient and robust learning / Self-supervised and contrastive
Similarity Maps for Self-Training Weakly-Supervised Phrase Grounding
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 |
|---|---|---|---|
| Fine-tune g using aggregated self-similarity maps as pseudo-labels | g · WSPG, VG, Flickr30k, and ReferIt; training on VG or MS-COCO | pointing game accuracy: 62.31 -> 66.63; 75.63 -> 79.95; 65.95 -> 70.25 | Helped |
| Fine-tune g using aggregated self-similarity maps as pseudo-labels | g · WWbL, VG, Flickr30k, and ReferIt; training on VG or MS-COCO | pointing game accuracy: 43.91 -> 45.90; 58.59 -> 62.98; 44.89 -> 45.14 | Helped |
| Use aggregated pseudo-label mask directly as the output, without fine-tuning g (ablation) | g · WSPG | pointing game accuracy: 62.31 -> 63.10 | Helped |
| Use aggregated pseudo-label mask directly as the output, without fine-tuning g (ablation) | g · WSPG | bounding box accuracy: 27.26 -> 24.07 | Mixed |
| Add pseudo-label supervision Lpseudo (ablation) | g++ · WSPG | pointing game accuracy: 62.31 -> 65.50 | Helped |
| Use Lpseudo + Lreg without Lfore or Lback (ablation) | g++ · WSPG | pointing game accuracy: 62.31 -> 59.40 | Hurt |
| Reduce the Lreg weight to one third when using Lpseudo (ablation) | g++ · WSPG | pointing game accuracy: 65.50 -> 65.80 | Helped |
| Use Lpseudo + Lfore + Lback without Lreg (ablation) | g++ · WSPG | pointing game accuracy: 65.50 -> 65.47 | Mixed |
| Select one highest-IoU similarity map instead of averaging relevant maps (ablation) | g++ · WSPG | pointing game accuracy: 66.63 -> 66.22 | Hurt |
| Replace the phrase-grounding encoder with a semantic segmentation encoder (ablation) | g++ · WSPG | pointing game accuracy: 62.31 -> 56.52 | Hurt |
| Replace the phrase-grounding encoder with an ImageNet classification encoder (ablation) | g++ · WSPG | pointing game accuracy: 62.31 -> 60.49 | Hurt |
Closest papers by meaning
- Momentum Pseudo-Labeling for Weakly Supervised Phrase Grounding (AAAI 2025 · similarity 0.90)
- Knowledge Aided Consistency for Weakly Supervised Phrase Grounding (CVPR 2018 · similarity 0.90)
- Improved Visual Grounding through Self-Consistent Explanations (CVPR 2024 · similarity 0.89)
- Siamese Learning with Joint Alignment and Regression for Weakly-Supervised Video Paragraph Grounding (CVPR 2024 · similarity 0.88)
- Omnipotent Distillation with LLMs for Weakly-Supervised Natural Language Video Localization: When Divergence Meets Consistency (AAAI 2024 · similarity 0.86)
- Learning Multi-Scale Video-Text Correspondence for Weakly Supervised Temporal Article Gronding (AAAI 2024 · similarity 0.86)
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.