Research map / Video / Action recognition
Self-Supervised Video Representation Learning With Meta-Contrastive Network
Yuanze Lin, Xun Guo, Yan Lu
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 MCN (meta-contrastive training) to the CMC baseline | CMC baseline R3D-18 · UCF101 fine-tune (pretrain UCF101) | top-1 accuracy (%): 78.6 -> 85.4 | Improved |
| add MCN to the CMC baseline | CMC baseline S3D · UCF101 fine-tune (pretrain UCF101) | top-1 accuracy (%): 76.7 -> 82.9 | Improved |
| add MCN to the CMC baseline | CMC baseline R3D-18 · HMDB51 fine-tune (pretrain UCF101) | top-1 accuracy (%): 47.1 -> 54.8 | Improved |
| add MCN to the CMC baseline | CMC baseline R3D-18 · UCF101 linear probe | top-1 accuracy (%): 64.6 -> 73.1 | Improved |
| add MCN to the CMC baseline | CMC baseline R3D-18 · HMDB51 linear probe | top-1 accuracy (%): 37.3 -> 42.9 | Improved |
| set meta loss weight alpha to 0.2 instead of 0.4 (ablation) | MCN R(2+1)D · UCF101 fine-tune (alpha sweep) | top-1 accuracy (%): 82.7 -> 84.8 | Improved |
| set alpha to 0.2 instead of 0.1 (ablation) | MCN R(2+1)D · UCF101 fine-tune (alpha sweep) | top-1 accuracy (%): 84.1 -> 84.8 | Improved |
| increase input frames from 16 to 64 (ablation) | MCN R(2+1)D · UCF101 fine-tune (input frames) | top-1 accuracy (%): 81.3 -> 86.7 | Improved |
| increase input frames from 16 to 64 (ablation) | Baseline R(2+1)D · UCF101 fine-tune (input frames) | top-1 accuracy (%): 74.6 -> 80.6 | Improved |
| add binary classification loss from the meta branch (ablation) | CMC baseline R(2+1)D · UCF101 fine-tune (component ablation) | top-1 accuracy (%): 77.3 -> 79.2 | Improved |
| add meta-train/meta-test stages (ablation) | CL + BL R(2+1)D · UCF101 fine-tune (component ablation) | top-1 accuracy (%): 79.2 -> 84.8 | Improved |
| pre-train on Kinetics-400 instead of UCF101 | MCN R3D · UCF101 fine-tune | top-1 accuracy (%): 85.4 -> 89.7 | Improved |
Closest papers by meaning
- Cycle-Contrast for Self-Supervised Video Representation Learning (NeurIPS 2020 · similarity 0.89)
- Cross-Architecture Self-Supervised Video Representation Learning (CVPR 2022 · similarity 0.88)
- Frame-Wise Action Representations for Long Videos via Sequence Contrastive Learning (CVPR 2022 · similarity 0.88)
- Self-Supervised Video Representation Learning via Latent Time Navigation (AAAI 2023 · similarity 0.88)
- Self-supervised Co-Training for Video Representation Learning (NeurIPS 2020 · similarity 0.88)
- Contrast and Order Representations for Video Self-Supervised Learning (ICCV 2021 · 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.
