Research map / Video / Video segmentation and tracking
VideoClusterNet: Self-Supervised and Adaptive Face Clustering for Videos
Devesh Walawalkar, Pablo Garrido
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 self-supervised finetuning of face ID model on the video's tracks (ablation) | ArcFace-R100 + HAC · MovieFaceCluster: The Hidden Soldier | cluster accuracy (%): 86.10 -> 91.52 | Improved |
| replace HAC with the proposed track clustering algorithm (cosine metric) (ablation) | VideoClusterNet · MovieFaceCluster: The Hidden Soldier | cluster accuracy (%): 91.52 -> 93.70 | Improved |
| use the SSL loss as similarity metric instead of cosine distance (ablation) | VideoClusterNet clustering · MovieFaceCluster: The Hidden Soldier | cluster accuracy (%): 93.70 -> 98.50 | Improved |
| use the SSL loss as similarity metric instead of Euclidean distance (ablation) | VideoClusterNet clustering · MovieFaceCluster: The Hidden Soldier | predicted cluster ratio: 3.5 (74/21) -> 1.04 (22/21) | Improved |
| add the finetuning method to different face ID base models (ablation) | FaRL, VGGFace2, ArcFace, AdaFace · MovieFaceCluster: The Hidden Soldier | cluster accuracy (%): 78.7 -> 90.2 (FaRL-P16); 86.1 -> 98.5 (ArcFace-R100) | Improved |
| apply VideoClusterNet vs prior methods on BBT episodes | BBT S01 · BBT S01 (combined) | WCP (%): 94.20 -> 98.70 | Improved |
| apply VideoClusterNet vs prior methods on BVS | BVS S05 · BVS S05 (combined) | WCP (%): 83.62 -> 96.10 | Improved |
| apply VideoClusterNet vs VCTRSF on hardest movies | MovieFaceCluster The Fortress · MovieFaceCluster: The Fortress | weighted cluster accuracy (%): 78.8 -> 89.3 | Improved |
| on BBT S1E2, VideoClusterNet vs VCTRSF | BBT S01E02 · BBT S01E02 | WCP (%): 99.84 vs 99.67 | No clear effect |
Closest papers by meaning
- Video Face Clustering With Unknown Number of Clusters (ICCV 2019 · similarity 0.91)
- End-To-End Face Detection and Cast Grouping in Movies Using Erdos-Renyi Clustering (ICCV 2017 · similarity 0.91)
- Clustering Millions of Faces by Identity (TPAMI 2017 · similarity 0.89)
- Self-supervised Video-centralised Transformer for Video Face Clustering (TPAMI 2023 · similarity 0.89)
- Automated Video Face Labelling for Films and TV Material (TPAMI 2018 · similarity 0.87)
- Linkage Based Face Clustering via Graph Convolution Network (CVPR 2019 · 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.
