Research map / Video / Action recognition
Building an Open-Vocabulary Video CLIP Model With Better Architectures, Optimization and Data
Zuxuan Wu, Zejia Weng, Wujian Peng, Xitong Yang, Ang Li, Larry Steven Davis, Yu–Gang Jiang
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 |
|---|---|---|---|
| add BLIP-2 + Llama-2 generated video captions as extra alignment loss (ablation) | Open-VCLIP++ · MSR-VTT text-to-video | Recall@1: 31.3 -> 34.6 | Improved |
| add generated captions as extra alignment loss (ablation) | Open-VCLIP++ · MSR-VTT text-to-video | Recall@10: 65.6 -> 69.5 | Improved |
| add generated captions as extra alignment loss (ablation) | Open-VCLIP++ · UCF-101 zero-shot | top-1 accuracy: 83.5 -> 84.1 | Improved |
| add generated captions as extra alignment loss (ablation) | Open-VCLIP++ · HMDB-51 zero-shot | top-1 accuracy: 53.2 -> 54.7 | Improved |
| add generated captions as extra alignment loss (ablation) | Open-VCLIP++ · Kinetics-600 zero-shot | top-1 accuracy: 73.0 -> 73.4 | Improved |
| standard fine-tuning on Kinetics-400 vs original CLIP | CLIP ViT-B/16 · Kinetics-600 zero-shot | top-1 accuracy: 68.1 -> 65.9 | Hurt |
| standard fine-tuning on Kinetics-400 vs original CLIP | CLIP ViT-L/14 · HMDB-51 zero-shot | top-1 accuracy: 55.0 -> 51.9 | Hurt |
| standard fine-tuning on Kinetics-400 vs original CLIP | CLIP ViT-B/16 · UCF-101 zero-shot | top-1 accuracy: 74.2 -> 79.7 | Improved |
| add interpolated weight regularization and SWA over fine-tuning | VCLIP ViT-B/16 · Kinetics-600 zero-shot | top-1 accuracy: 65.9 -> 73.4 | Improved |
| add interpolated weight regularization and SWA over fine-tuning | VCLIP ViT-B/16 · HMDB-51 zero-shot | top-1 accuracy: 49.2 -> 54.7 | Improved |
| add IWR to SWA (use both) (ablation) | Open-VCLIP++ · zero-shot UCF, HMDB, K600 | top-1 accuracy | Improved |
| use IWR instead of l2 weight penalty (ablation) | Open-VCLIP · UCF, HMDB, K600 | zero-shot and close-set accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ST-Adapter — ST-Adapter: Parameter-Efficient Image-to-Video Transfer Learning (NeurIPS 2022) · better on 1 of 1 shared result
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
- FROSTER: Frozen CLIP is A Strong Teacher for Open-Vocabulary Action Recognition (ICLR 2024 · similarity 0.89)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.89)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.89)
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation (ICLR 2024 · similarity 0.88)
- ViLT-CLIP: Video and Language Tuning CLIP with Multimodal Prompt Learning and Scenario-Guided Optimization (AAAI 2024 · similarity 0.88)
- TPR: Topology-Preserving Reservoirs for Generalized Zero-Shot Learning (NeurIPS 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.
