Research map / Video / Action recognition
Temporal Alignment Networks for Long-Term Video
Check 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 long-context transformers trained on HTM-370K | MIL-NCE visual and textual features · HTM-Align | Recall@1: 31.3 -> 45.8 | Helped |
| Add co-training with mutual-agreement denoising (ablation) | TAN initialized in stage 1 · HTM-Align | Recall@1: 45.8 -> 49.4 | Helped |
| Use 3 transformer layers instead of 6 (ablation) | TAN and dual encoder · HTM-Align | Recall@1: 45.8 -> 42.3 | Hurt |
| Set alignability threshold alpha to 0.25 rather than 0.5 (ablation) | TAN co-training · HTM-Align | Recall@1: 49.4 -> 42.5 | Hurt |
| Set alignability threshold alpha to 0.75 rather than 0.5 (ablation) | TAN co-training · HTM-Align | Recall@1: 49.4 -> 48.8 | No clear effect |
| Increase input video window from 32 to 64 seconds (ablation) | TAN · HTM-Align | Recall@1: 41.1 -> 49.4 | Helped |
| Increase input video window from 64 to 128 seconds (ablation) | TAN · HTM-Align | Recall@1: 49.4 -> 48.4 | No clear effect |
| Use HTM-Full rather than HTM-370K for co-training (ablation) | TAN · HTM-Align | Recall@1: 49.4 -> 49.2 | No clear effect |
| Add co-training stage after initialization (ablation) | TAN dual encoder for retrieval · YouCook2 (zero-shot) | R@1: 16.8 -> 20.1 | Helped |
| Fine-tune S3D using TAN auto-aligned HTM text-video pairs | S3D backbone · UCF101 (linear probe) | accuracy: 82.1 -> 83.2 | Helped |
| Fine-tune TAN on Breakfast with soft-DTW loss | TAN (S1+S2) · Breakfast-Action | F-Acc: 65.1 -> 68.3 | Helped |
Closest papers by meaning
- VidLA: Video-Language Alignment at Scale (CVPR 2024 · similarity 0.90)
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · similarity 0.90)
- Temporal Alignment-Free Video Matching for Few-shot Action Recognition (CVPR 2025 · similarity 0.89)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.88)
- Self-alignment of Large Video Language Models with Refined Regularized Preference Optimization (NeurIPS 2025 · similarity 0.88)
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 2024 · 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.