Research map / Vision-language and multimodal / CLIP and zero-shot
Probabilistic Prompt Learning for Dense Prediction
Hyeongjun Kwon, Taeyong Song, Somi Jeong, Jin Kim, Jinhyun Jang, Kwanghoon Sohn
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 |
|---|---|---|---|
| Use probabilistic, visual-context-aware prompts instead of deterministic prompts | Semantic FPN · ADE20k val | mIoU-SS: 43.5 -> 44.7 | Improved |
| Use probabilistic, visual-context-aware prompts instead of deterministic prompts | Mask R-CNN, ResNet-50 · COCO val | AP^b: 40.2 -> 41.0 | Improved |
| Use probabilistic, visual-context-aware prompts instead of deterministic prompts | Mask R-CNN, ResNet-50 · COCO val | AP^m: 37.6 -> 38.3 | Improved |
| Model prompt distributions with visual-text context rather than text-only prompt statistics | Semantic FPN, ResNet-50 · ADE20k val | mIoU-SS: 42.6 -> 44.7 | Improved |
| Add diversity loss (ablation) | PPL · ADE20k val | mIoU: 42.0 -> 42.4 | Improved |
| Add KL divergence loss (ablation) | PPL · ADE20k val | mIoU: 42.0 -> 43.6 | Improved |
| Add diversity loss and KL divergence loss together (ablation) | PPL · ADE20k val | mIoU: 42.0 -> 44.3 | Improved |
| Replace pixel-matching loss with probabilistic pixel-text matching loss (ablation) | PPL with diversity and KL losses · ADE20k val | mIoU: 44.3 -> 44.7 | Improved |
| Use 3 attribute prompts instead of 1 (ablation) | PPL · ADE20k val | mIoU: 43.8 -> 44.7 | Improved |
| Increase attribute prompts from 3 to 5 (ablation) | PPL · ADE20k val | mIoU: 44.7 -> 44.2 | Hurt |
| Set KL-divergence weight β to 10⁻⁷ instead of 10⁻⁵ (ablation) | PPL · ADE20k val | mIoU: 44.7 -> 42.4 | Hurt |
| Set KL-divergence weight β to 10⁻⁴ instead of 10⁻⁵ (ablation) | PPL · ADE20k val | mIoU: 44.7 -> 44.1 | Hurt |
Closest papers by meaning
- ProAPO: Progressively Automatic Prompt Optimization for Visual Classification (CVPR 2025 · similarity 0.88)
- Tree of Attributes Prompt Learning for Vision-Language Models (ICLR 2025 · similarity 0.88)
- Prompt Tuning In a Compact Attribute Space (AAAI 2025 · similarity 0.88)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.87)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.87)
- Make Prompts Adaptable: Bayesian Modeling for Vision-Language Prompt Learning with Data-Dependent Prior (AAAI 2024 · 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.
