Research map / Video / Action recognition
Learning From Video and Text via Large-Scale Discriminative Clustering
Antoine Miech, Jean-Baptiste Alayrac, Piotr Bojanowski, Ivan Laptev, Josef Sivic
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 the joint person-action constraint (ablation) | Full setup · Action recognition (5 movies, 13 classes) | mAP (%): 12.5 -> 16.4 | Improved |
| train on 66 movies instead of only the 5 evaluated movies (ablation) | Full setup · Action recognition (5 movies, 13 classes) | mAP (%): 12.4 -> 16.4 | Improved |
| use joint person-action constraint for interacting actions (Fight, Hand Shake, Hug, Kiss) (ablation) | Full setup · Action recognition (5 movies, 13 classes) | per-class AP: Fight 12.6 -> 5.9 | Hurt |
| add weak script supervision with discriminative clustering over fully supervised small set (ablation) | Full setup · Action recognition (5 movies, 13 classes) | mAP (%): 9.3 -> 16.4 | Improved |
| add background constraint alpha=0.3 instead of 0 (ablation) | Our method · Casablanca person recognition | accuracy (%): 58 -> 82 | Improved |
| raise background constraint alpha to 1.0 (ablation) | Our method · Casablanca person recognition | accuracy (%): 84 -> 55 | Hurt |
| add background constraint beta (ablation) | Full setup · Action recognition (beta sweep) | mAP (%): 15.0 -> 16.6 | No clear effect |
| use our discriminative-clustering person model over Parkhi et al. baseline | Our method · Casablanca person recognition | accuracy (%): 74% -> 83% | Improved |
| use our model on Buffy episode 1 | Ours · Buffy season 5 episode 1 | AP (%): 99 -> 98 | Hurt |
| use our model on Buffy episode 3 | Ours · Buffy season 5 episode 3 | AP (%): 94 -> 98 | Improved |
Closest papers by meaning
- Large-Scale Weakly-Supervised Pre-Training for Video Action Recognition (CVPR 2019 · similarity 0.87)
- A flexible model for training action localization with varying levels of supervision (NeurIPS 2018 · similarity 0.87)
- Uncertainty-Aware Weakly Supervised Action Detection from Untrimmed Videos (ECCV 2020 · similarity 0.86)
- Revisiting Foreground and Background Separation in Weakly-supervised Temporal Action Localization: A Clustering-based Approach (ICCV 2023 · similarity 0.86)
- Unsupervised Action Discovery and Localization in Videos (ICCV 2017 · similarity 0.85)
- Action Sets: Weakly Supervised Action Segmentation Without Ordering Constraints (CVPR 2018 · 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.
