Research map / Video / Action recognition
Rethinking Learning Approaches for Long-Term Action Anticipation
Megha Nawhal, Akash Abdu Jyothi, Greg Mori
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 L1 temporal loss to IoU loss (ablation) | Anticipatr · Breakfast (observe 20%, anticipate 50%) | mean-over-classes accuracy: 26.4 -> 28.6 | Improved |
| add IoU temporal loss to L1 loss (ablation) | Anticipatr · Breakfast (observe 20%, anticipate 50%) | mean-over-classes accuracy: 28.2 -> 28.6 | Improved |
| add L1 temporal loss to IoU loss (ablation) | Anticipatr · EK-55 (all classes) | mAP: 34.9 -> 39.1 | Improved |
| increase anticipation queries from 50 to 150 (ablation) | Anticipatr · Breakfast (observe 20%, anticipate 50%) | mean-over-classes accuracy: 24.3 -> 28.6 | Improved |
| increase anticipation queries from 150 to 500 (ablation) | Anticipatr · Breakfast (observe 20%, anticipate 50%) | mean-over-classes accuracy: 28.6 -> 27.3 | Hurt |
| increase anticipation queries from 900 to 2700 (ablation) | Anticipatr · EK-55 (all classes) | mAP: 39.1 -> 38.2 | Hurt |
| use segment window 16 instead of 4 (ablation) | Anticipatr · Breakfast (observe 20%, anticipate 50%) | mean-over-classes accuracy: 26.1 -> 28.6 | Improved |
| use segment window 16 instead of 64 (ablation) | Anticipatr · Breakfast (observe 20%, anticipate 50%) | mean-over-classes accuracy: 28.1 -> 28.6 | Improved |
| add segment encoder (segment-level representations) (ablation) | Anticipatr · Breakfast, 50Salads, EK-55, EGTEA+ | accuracy / mAP | Improved |
| train segment encoder in a separate first stage (ablation) | Anticipatr · Breakfast, 50Salads, EK-55, EGTEA+ | accuracy / mAP | Improved |
| fine-tune the segment encoder during anticipation stage (ablation) | Anticipatr · Breakfast, 50Salads, EK-55, EGTEA+ | accuracy / mAP | Hurt |
| predict a set of action instances instead of per-time-instant labels (ablation) | Anticipatr · Breakfast, 50Salads, EK-55, EGTEA+ | accuracy / mAP | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- EGO-TOPO — Ego-Topo: Environment Affordances From Egocentric Video (CVPR 2020) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Future Transformer for Long-Term Action Anticipation (CVPR 2022 · similarity 0.91)
- Time-Conditioned Action Anticipation in One Shot (CVPR 2019 · similarity 0.91)
- ActFusion: a Unified Diffusion Model for Action Segmentation and Anticipation (NeurIPS 2024 · similarity 0.90)
- Towards Understanding Future: Consistency Guided Probabilistic Modeling for Action Anticipation (AAAI 2024 · similarity 0.89)
- AntGPT: Can Large Language Models Help Long-term Action Anticipation from Videos? (ICLR 2024 · similarity 0.89)
- Can't Make an Omelette Without Breaking Some Eggs: Plausible Action Anticipation Using Large Video-Language Models (CVPR 2024 · similarity 0.87)
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.
