Research map / Video / Action recognition
Title Learning Latent Subevents in Activity Videos Using Temporal Attention Filters
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 learned static temporal filters | TDD activity recognition · DogCentric | accuracy: 76.6 % -> 79.6 % | Helped |
| Add LSTM-based dynamic adjustment of temporal filters | TDD activity recognition with temporal filters · DogCentric | accuracy: 79.6 % -> 81.4 % | Helped |
| Add per-activity learned temporal filters | TDD activity recognition · HMDB | accuracy: 58.9 % -> 68.4 % | Helped |
| Replace max pooling with learned temporal filters | C3D activity recognition · HMDB | accuracy: 48.5 % -> 57.7 % | Helped |
| Replace max pooling with learned temporal filters | TDD activity recognition · HMDB | accuracy: 57.1 % -> 68.4 % | Helped |
| Use fixed temporal pyramid filters instead of learned filters (ablation) | VGG activity recognition · HMDB (shared filters) | accuracy: 41.56 % -> 42.50 % | Hurt |
| Use learned filters with N=1 rather than N=3 (ablation) | VGG activity recognition · HMDB (shared filters) | accuracy: 41.23 % -> 42.50 % | Hurt |
| Use learned filters with N=1 rather than N=3 (ablation) | C3D activity recognition · HMDB (shared filters) | accuracy: 50.00 % -> 50.35 % | Helped |
| Use learned filters with N=1 rather than N=3 (ablation) | TDD activity recognition · HMDB (shared filters) | accuracy: 58.87 % -> 59.03 % | Hurt |
| Use LSTM dynamic filters with three filters rather than one (ablation) | VGG activity recognition · HMDB (shared filters) | accuracy: 42.72 % -> 43.03 % | Helped |
| Use LSTM dynamic filters with three filters rather than one (ablation) | C3D activity recognition · HMDB (shared filters) | accuracy: 51.20 % -> 49.81 % | Hurt |
| Use LSTM dynamic filters with three filters rather than one (ablation) | TDD activity recognition · HMDB (shared filters) | accuracy: 58.04 % -> 58.93 % | Helped |
Closest papers by meaning
- Learning Latent Super-Events to Detect Multiple Activities in Videos (CVPR 2018 · similarity 0.92)
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.88)
- Coarse-Fine Networks for Temporal Activity Detection in Videos (CVPR 2021 · similarity 0.87)
- Segregated Temporal Assembly Recurrent Networks for Weakly Supervised Multiple Action Detection (AAAI 2019 · similarity 0.87)
- Shrinking Temporal Attention in Transformers for Video Action Recognition (AAAI 2022 · similarity 0.87)
- Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders (NeurIPS 2025 · 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.