Research map / Video / Video segmentation and tracking
GOCA: Guided Online Cluster Assignment for Self-Supervised Video Representation Learning
Huseyin Coskun, Alireza Zareian, Joshua L. Moore, Federico Tombari, Chen 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 guided cluster assignment (flow and RGB priors) instead of separate training (ablation) | Sep · UCF retrieval (S3D, UCF-pretrained) | R@1: 65.0 -> 70.8 | Improved |
| use guided assignment instead of naive feature averaging (ablation) | Avg · UCF retrieval (S3D, UCF-pretrained) | R@1: 50.8 -> 70.8 | Improved |
| use guided assignment instead of enforcing RGB/flow feature similarity (ablation) | VICC · UCF retrieval, flow test (S3D) | R@1: 59.7 -> 67.9 | Improved |
| use guided assignment instead of enforcing RGB/flow feature similarity (ablation) | VICC · UCF retrieval, RGB test (S3D) | R@1: 62.1 -> 63.4 | Improved |
| add joint training with guidance to RGB-only model (ablation) | SView RGB · UCF retrieval (S3D) | R@1: 41.3 -> 63.4 | Improved |
| add prototype regularization (ablation) | SView RGB · UCF retrieval (S3D) | R@1: 40.9 -> 42.3 | Improved |
| add prototype regularization (ablation) | GOCA+ · UCF retrieval (S3D) | R@1: 69.1 -> 70.8 | Improved |
| add prototype regularization (ablation) | GOCA+ · HMDB retrieval (S3D) | R@1: 32.9 -> 33.6 | Improved |
| use lambda1 0.02 and lambda2 0.03 versus lambda1 0.01 and lambda2 0.01 (ablation) | GOCA · UCF RGB-only val (200 epochs) | R@1: 56.4 -> 60.5 | Improved |
| use lambda2 0.04 versus 0.03 at lambda1 0.02 (ablation) | GOCA · UCF RGB-only val (200 epochs) | R@1: 60.9 -> 58.5 | Hurt |
| add optical flow as guidance prior (GOCA over single-view RGB) (ablation) | SView RGB · UCF cluster accuracy | clustering accuracy: 42.1 -> 57.3 | Improved |
| combine RGB and flow at test time | GOCA · UCF linear classification (S3D, UCF-pretrained) | top-1 accuracy: 69.2 -> 81.1 | Improved |
Closest papers by meaning
- Semi-Supervised Learning Made Simple With Self-Supervised Clustering (CVPR 2023 · similarity 0.88)
- ClusterFit: Improving Generalization of Visual Representations (CVPR 2020 · similarity 0.88)
- Multimodal Clustering Networks for Self-Supervised Learning From Unlabeled Videos (ICCV 2021 · similarity 0.88)
- Efficient Unsupervised Visual Representation Learning with Explicit Cluster Balancing (ECCV 2024 · similarity 0.87)
- Partial Multi-View Clustering via Self-Supervised Network (AAAI 2024 · similarity 0.86)
- Investigating and Mitigating the Side Effects of Noisy Views for Self-Supervised Clustering Algorithms in Practical Multi-View Scenarios (CVPR 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.
