Research map / Video / Action recognition
Task Graph Maximum Likelihood Estimation for Procedural Activity Understanding in Egocentric Videos
Luigi Seminara, Giovanni Maria Farinella, Antonino Furnari
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 |
|---|---|---|---|
| learn task graph by gradient-based maximum likelihood (DO) instead of MSG2 heuristics | task graph generation · CaptainCook4D | F1: 73.3 -> 87.8 | Improved |
| DO instead of Count-Based | task graph generation · EgoPER | F1: 80.8 -> 91.0 | Improved |
| DO instead of Llama-3.1-405B-Instruct | task graph generation · EgoProceL | F1: 58.7 -> 72.3 | Improved |
| use feature-based TGT-text instead of direct adjacency optimization (DO) | task graph generation · CaptainCook4D | F1: 87.8 -> 80.8 | Hurt |
| use TGT-text instead of DO for downstream Ego-Exo4D optional keystep (instance-level) | procedure understanding · Ego-Exo4D Val | Optional keysteps: 74.52 -> 75.56 | Mixed |
| use DO instead of TGT-text for procedural mistakes (instance-level) | procedure understanding · Ego-Exo4D Val | Proc. Mistakes: 78.83 -> 84.52 | Improved |
| learned graph (TGT-text, keystep prediction) instead of graph-based baseline at procedure level | procedure understanding · Ego-Exo4D Val | Prev. Keysteps: 64.56 -> 71.37 | Improved |
| use explicit DO task graph instead of implicit PREGO for mistake detection (GT actions) | online mistake detection · Assembly101-O | average F1: 39.4 -> 75.9 | Improved |
| use DO instead of Count-Based on EPIC-Tent-O (GT actions) | online mistake detection · EPIC-Tent-O | average F1: 57.7 -> 58.3 | Improved |
| use TGT-text instead of DO with more descriptive step names | online mistake detection · EPIC-Tent-O | average F1: 58.3 -> 64.1 | Improved |
| replace ground-truth actions with predicted actions for DO | online mistake detection · Assembly101-O | average F1: 75.9 -> 53.5 | Hurt |
| DO mistake-class F1 with predicted actions vs MSG2 | online mistake detection · Assembly101-O | Mistake F1: 33.2 -> 28.1 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PREGO — PREGO: Online Mistake Detection in PRocedural EGOcentric Videos (CVPR 2024) · better on 2 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
- Differentiable Task Graph Learning: Procedural Activity Representation and Online Mistake Detection from Egocentric Videos (NeurIPS 2024 · similarity 0.96)
- Error Detection in Egocentric Procedural Task Videos (CVPR 2024 · similarity 0.90)
- PHGC: Procedural Heterogeneous Graph Completion for Natural Language Task Verification in Egocentric Videos (CVPR 2025 · similarity 0.89)
- Progress-Aware Online Action Segmentation for Egocentric Procedural Task Videos (CVPR 2024 · similarity 0.87)
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.87)
- CaptainCook4D: A Dataset for Understanding Errors in Procedural Activities (NeurIPS 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.
