Research map / Video / Action recognition
Unsupervised Learning From Video With Deep Neural Embeddings
Check 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 |
|---|---|---|---|
| add a static pathway to the SlowFast dynamic pathway | VIE-SlowFast · UCF101 and HMDB51 | top-1 accuracy: 77.0 -> 78.2; 46.5 -> 50.5 | Helped |
| add a static pathway to the SlowFast dynamic pathway | VIE-SlowFast · ImageNet validation | top-1 transfer accuracy (%): 36.60 -> 43.69 | Helped |
| use multi-frame dynamic processing instead of single-frame processing | VIE · Kinetics action recognition | top-1 transfer accuracy (%): Multi-frame models substantially outperform single-frame versions. | Helped |
| use multi-frame dynamic processing instead of single-frame processing | VIE · ImageNet validation | top-1 transfer accuracy (%): VIE-Single 42.33; VIE-3DResNet 32.62 | Hurt |
| replace Local Aggregation with Instance Recognition loss (ablation) | VIE-Single · Kinetics validation, Conv5 | top-1 transfer accuracy (%): 43.85 -> 44.41 | Hurt |
| train on more Kinetics videos (ablation) | VIE-Single · ImageNet validation, Conv5 | top-1 transfer accuracy (%): 39.85; 36.23; 42.33 | Helped |
| use full videos rather than temporally clipped videos (ablation) | VIE-Single · ImageNet validation, Conv5 | top-1 transfer accuracy (%): 39.36; 37.83; 42.33 | Helped |
| use color-noise augmentation during fine-tuning (ablation) | VIE · UCF101 | top-1 accuracy: 72.3 -> 75.5 | Helped |
| use color-noise augmentation during fine-tuning (ablation) | VIE-TwoPathway-SF · HMDB51 | top-1 accuracy: 50.5 -> 52.5 | Mixed |
| use the complete video rather than splitting it into temporal bins (ablation) | VIE-Single · Kinetics and ImageNet transfer | top-1 accuracy (%): Kinetics Conv5: 44.41, 44.24, 43.33; ImageNet Conv5: 42.33, 39.36, 37.83 | Mixed |
| use additional videos for unsupervised training (ablation) | VIE-Single · ImageNet validation | top-1 transfer accuracy (%): 39.85; 36.23; 42.33 | Helped |
| replace learned multi-frame processing with single-frame processing over the same sampled inputs (ablation) | VIE multi-frame models · Kinetics validation, Conv5 | top-1 transfer accuracy (%): 44.27; 44.62; 44.29 | Helped |
Closest papers by meaning
- VONet: Unsupervised Video Object Learning With Parallel U-Net Attention and Object-wise Sequential VAE (ICLR 2024 · similarity 0.87)
- Unsupervised Learning of Action Classes With Continuous Temporal Embedding (CVPR 2019 · similarity 0.87)
- Unsupervised Deep Learning of Mid-Level Video Representation for Action Recognition (AAAI 2018 · similarity 0.87)
- Self-Supervised Video Representation Learning With Odd-One-Out Networks (CVPR 2017 · similarity 0.87)
- Unsupervised Learning of View-invariant Action Representations (NeurIPS 2018 · similarity 0.87)
- Unsupervised Video Domain Adaptation with Masked Pre-Training and Collaborative Self-Training (CVPR 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.