Research map / Video / Action recognition
Low-Fidelity Video Encoder Optimization for Temporal Action Localization
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 LoFi end-to-end video encoder optimization with TAL loss (Long C-LoFi) after classification pre-training (ablation) | G-TAD with TSM-R18 · ActivityNet-v1.3 | average mAP: 33.59 -> 34.49 | Helped |
| add LoFi end-to-end video encoder optimization (ablation) | G-TAD with TSM-R18 · HACS-v1.1 | average mAP: 23.00 -> 24.64 | Helped |
| fine-tune encoder with action classification on target positive segments (ACP+) instead of LoFi (ablation) | G-TAD with TSM-R18 · ActivityNet-v1.3 | average mAP: 33.59 -> 33.84 | No clear effect |
| replace Kinetics action pre-training with ImageNet image pre-training (ablation) | G-TAD with TSM-R18 · HACS-v1.1 | average mAP: 23.00 -> 20.18 | Hurt |
| use spatial-only low fidelity (S-LoFi) rather than standard training (ablation) | G-TAD with TSM-R18 · ActivityNet-v1.3 | average mAP: 33.59 -> 34.12 | Helped |
| use temporal-only low fidelity (T-LoFi) (ablation) | G-TAD with TSM-R18 · ActivityNet-v1.3 | average mAP: 33.59 -> 34.32 | Helped |
| cycle through low-fidelity configurations (Long C-LoFi) rather than single configuration (ablation) | LoFi · ActivityNet-v1.3 | average mAP: 34.32 -> 34.49 | Helped |
| add LoFi to R(2+1)D-R18 encoder (ablation) | G-TAD · ActivityNet-v1.3 | average mAP: 33.55 -> 34.21 | Helped |
| add LoFi to deeper TSM-R50 encoder (ablation) | G-TAD · ActivityNet-v1.3 | average mAP: 34.26 -> 34.96 | Helped |
| add self-supervised pre-training (Arrow of Time) on top of ACP | G-TAD with TSM-R18 · ActivityNet-v1.3 | average mAP: 33.59 -> 33.72 | No clear effect |
| add SpeedNet self-supervision on top of ACP | G-TAD with TSM-R18 · ActivityNet-v1.3 | average mAP: 33.59 -> 33.75 | No clear effect |
Closest papers by meaning
- Temporal Action Localization with Cross Layer Task Decoupling and Refinement (AAAI 2025 · similarity 0.91)
- Test-Time Zero-Shot Temporal Action Localization (CVPR 2024 · similarity 0.89)
- Energy vs. Noise: Towards Robust Temporal Action Localization in Open-World (AAAI 2025 · similarity 0.88)
- Realigning Confidence with Temporal Saliency Information for Point-Level Weakly-Supervised Temporal Action Localization (CVPR 2024 · similarity 0.87)
- Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer (CVPR 2025 · similarity 0.87)
- Masked Diffusion Vision-Language Models for Temporal Action Localization (NeurIPS 2026 · 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.