Research map / Label-efficient and robust learning / Domain adaptation and generalization
AutoLabel: CLIP-Based Framework for Open-Set Video Domain Adaptation
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 |
|---|---|---|---|
| Replace ResNet101 with CLIP in CEVT | CEVT · HMDB→UCF and UCF→HMDB | HOS: +3.9 and +3.1 points | Helped |
| Use AutoLabel candidate target-private labels instead of threshold-based unknown rejection | ActionCLIP · HMDB→UCF and UCF→HMDB | HOS: +0.8 and +2.6 points | Helped |
| Use cluster-level compositional candidate labels instead of per-instance detected object labels | ActionCLIP-ZOC · HMDB→UCF and UCF→HMDB | HOS: +12.5 and +2.1 points | Helped |
| Use AutoLabel candidate labels instead of oracle target-private names | ActionCLIP · Epic-Kitchens | HOS: 38.2 vs 33.5 | Helped |
| Use 35-45 target clusters (ablation) | AutoLabel · HMDB→UCF and Epic-Kitchens D1→D2 | HOS: 35-45 | Helped |
| Increase cluster granularity beyond the ground-truth target-category count (ablation) | AutoLabel · HMDB→UCF | HOS | Helped |
| Use candidate labels with 2-3 attribute tokens rather than many tokens (ablation) | AutoLabel · HMDB↔UCF and Epic-Kitchens | HOS: 2-3 | Helped |
| Increase the number of attribute tokens substantially (ablation) | AutoLabel · HMDB↔UCF and Epic-Kitchens | HOS | Hurt |
| Use around 20% of the most confident target pseudo-labels for fine-tuning (ablation) | AutoLabel · HMDB→UCF and Epic-Kitchens D1→D2 | HOS: 20% | Helped |
| Add pseudo-labels beyond the most confident 20% (ablation) | AutoLabel · HMDB→UCF and Epic-Kitchens D1→D2 | HOS: 20% | Hurt |
| Vary the attribute-matching threshold (ablation) | AutoLabel · HMDB→UCF and Epic-Kitchens D1→D2 | HOS: 80% and 40% | No clear effect |
Closest papers by meaning
- Unsupervised Video Domain Adaptation with Masked Pre-Training and Collaborative Self-Training (CVPR 2024 · similarity 0.87)
- Unknown Prompt the only Lacuna: Unveiling CLIP's Potential for Open Domain Generalization (CVPR 2024 · similarity 0.85)
- Learning to Adapt Frozen CLIP for Few-Shot Test-Time Domain Adaptation (ICLR 2025 · similarity 0.85)
- Image-to-video Adaptation with Outlier Modeling and Robust Self-learning (AAAI 2025 · similarity 0.85)
- OSLoPrompt: Bridging Low-Supervision Challenges and Open-Set Domain Generalization in CLIP (CVPR 2025 · similarity 0.84)
- Source-Free Domain Adaptation with Frozen Multimodal Foundation Model (CVPR 2024 · similarity 0.84)
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.