Research map / Video / Action recognition
Self-Supervised Video Representation Learning With Odd-One-Out Networks
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 |
|---|---|---|---|
| Pretrain with odd-one-out learning using Stack-of-Diff. clips | AlexNet · UCF101 split 1 | accuracy: 51.3 -> 60.3 | Helped |
| Pretrain with odd-one-out learning using Stack-of-Diff. clips | AlexNet · HMDB51 split 1 | accuracy: 28.3 -> 32.5 | Helped |
| Use random sampling instead of consecutive sampling (ablation) | O3N · UCF101 | accuracy: 50.6 -> 53.2 | Helped |
| Reduce fully connected activations from 4096 to 128 (ablation) | O3N with SOD fusion · UCF101 | mean accuracy: 51.3 -> 53.2 | Helped |
| Replace SOD fusion with concatenation (ablation) | O3N with 128 activations · UCF101 | mean accuracy: 53.2 -> 50.1 | Hurt |
| Use concatenation instead of SOD fusion (ablation) | O3N with 128 activations · UCF101 split 1 self-supervised validation | accuracy: 29.6 -> 33.6 | Helped |
| Increase question size from 2 to 6 elements (ablation) | O3N · UCF101 | mean accuracy: 49.3 -> 53.2 | Helped |
| Increase question size from 6 to 10 elements (ablation) | O3N · UCF101 | mean accuracy: 53.2 -> 52.8 | Hurt |
| Use a 2-element rather than a 6-element question (ablation) | O3N · UCF101 split 1 self-supervised validation | accuracy: 29.6 -> 73.0 | Helped |
| Use Stack-of-Diff. rather than Dynamic Image encoding (ablation) | O3N · UCF101 | accuracy: 53.2 -> 60.0 | Helped |
| Use Stack-of-Diff. rather than Dynamic Image encoding (ablation) | O3N · HMDB51 | accuracy: 26.0 -> 32.4 | Helped |
| Use O3N pretraining rather than random initialization | AlexNet with Sum-of-diff. encoding · UCF101 | accuracy: 43.4 -> 54.3 | Helped |
Closest papers by meaning
- Self-Supervised Learning of Compressed Video Representations (ICLR 2021 · similarity 0.87)
- Unsupervised Learning From Video With Deep Neural Embeddings (CVPR 2020 · similarity 0.87)
- Learning by Aligning Videos in Time (CVPR 2021 · similarity 0.87)
- TimeBalance: Temporally-Invariant and Temporally-Distinctive Video Representations for Semi-Supervised Action Recognition (CVPR 2023 · similarity 0.87)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.86)
- Is ImageNet worth 1 video? Learning strong image encoders from 1 long unlabelled video (ICLR 2024 · similarity 0.85)
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.