Research map / Video / Action recognition
Contrastive Learning of Image Representations With Cross-Video Cycle-Consistency
Haiping Wu, Xiaolong Wang
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 cross-video cycle loss to intra-image + intra-video loss (ablation) | contrastive pretraining (ResNet-18) · OTB-100 tracking | precision: 60.1 -> 65.6 | Improved |
| add cross-video cycle loss to intra-image + intra-video loss (ablation) | contrastive pretraining (ResNet-50) · OTB-100 tracking | precision: 68.4 -> 69.3 | Improved |
| add intra-video invariance to intra-image loss (ablation) | contrastive pretraining (ResNet-18) · OTB-100 tracking | precision: 53.7 -> 60.1 | Improved |
| add intra-video invariance to intra-image loss (ablation) | contrastive pretraining (ResNet-50) · OTB-100 tracking | precision: 47.4 -> 68.4 | Improved |
| add intra-video invariance to intra-image loss (ablation) | contrastive pretraining (ResNet-18) · ImageNet linear probe | top-1 accuracy: 33.0 -> 33.1 | No clear effect |
| add cross-video cycle loss to intra-image + intra-video loss (ablation) | contrastive pretraining (ResNet-18) · ImageNet linear probe | top-1 accuracy: 33.1 -> 34.4 | Improved |
| add cross-video cycle loss to intra-image + intra-video loss (ablation) | contrastive pretraining (ResNet-50) · ImageNet linear probe | top-1 accuracy: 55.1 -> 55.6 | Improved |
| add intra-video invariance to intra-image loss (ablation) | contrastive pretraining (ResNet-18) · UCF101 action recognition | top-1 accuracy: 45.1 -> 48.4 | Improved |
| add cross-video cycle loss to intra-image + intra-video loss (ablation) | contrastive pretraining (ResNet-18) · UCF101 action recognition | top-1 accuracy: 48.4 -> 50.5 | Improved |
| add cross-video cycle loss to intra-image + intra-video loss (ablation) | contrastive pretraining (ResNet-50) · UCF101 action recognition | top-1 accuracy: 66.2 -> 67.1 | Improved |
| add intra-video loss to cycle loss alone (ablation) | cross-video cycle objective · ImageNet-100 linear probe | top-1 accuracy: 45.78 -> 58.50 | Improved |
| enlarge neighbor set from 256 to 16384 (ablation) | full loss, ResNet-18 · ImageNet-100 linear probe | top-1 accuracy: 56.68 -> 58.48 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SeCo — SeCo: Exploring Sequence Supervision for Unsupervised Representation Learning (AAAI 2021) · better on 0 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
- Cycle-Contrast for Self-Supervised Video Representation Learning (NeurIPS 2020 · similarity 0.90)
- Contrast and Order Representations for Video Self-Supervised Learning (ICCV 2021 · similarity 0.89)
- Self-Supervised Video Representation Learning in a Heuristic Decoupled Perspective (IJCV 2026 · similarity 0.89)
- Delving into Inter-Image Invariance for Unsupervised Visual Representations (IJCV 2022 · similarity 0.89)
- CrossCLR: Cross-Modal Contrastive Learning for Multi-Modal Video Representations (ICCV 2021 · similarity 0.88)
- A Broad Study on the Transferability of Visual Representations With Contrastive 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.
