Research map / Video / Video-language understanding
MOMA-LRG: Language-Refined Graphs for Multi-Object Multi-Actor Activity Parsing
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 Kinetics-400 pre-training (ablation) | Swin-B · MOMA-LRG activity classification | acc@1: 0.5282 -> 0.8576 | Helped |
| add Kinetics-400 pre-training (ablation) | MViT-B · MOMA-LRG activity classification | acc@1: 0.5140 -> 0.7731 | Helped |
| add Kinetics-400 pre-training (ablation) | SlowFast-R50 · MOMA-LRG activity classification | acc@1: 0.4375 -> 0.7569 | Helped |
| add Kinetics-400 pre-training (ablation) | Swin-B · MOMA-LRG sub-activity classification | acc@1: 0.3817 -> 0.6450 | Helped |
| add Kinetics-400 pre-training (ablation) | MViT-B · MOMA-LRG sub-activity classification | acc@1: 0.4375 -> 0.6032 | Helped |
| add Kinetics-400 pre-training (ablation) | SlowFast-R50 · MOMA-LRG sub-activity classification | acc@1: 0.3739 -> 0.5625 | Helped |
| use oracle ground-truth bounding boxes (ablation) | GCN · MOMA-LRG activity classification | acc@1: 0.7837 -> 0.9502 | Helped |
| use oracle ground-truth bounding boxes (ablation) | GCN · MOMA-LRG sub-activity classification | acc@1: 0.3829 -> 0.563 | Helped |
| add text stream with contrastive loss (ablation) | MViT-B · MOMA-LRG spatio-temporal atomic action segmentation | score: 0.2130 -> 0.2353 | Helped |
| add text stream with contrastive loss (ablation) | SlowFast-R50 · MOMA-LRG spatio-temporal atomic action segmentation | score: 0.1975 -> 0.2023 | Helped |
| use ground-truth boxes and classes (PredCls) instead of detected (SGDet) | RelDN scene graph · MOMA-LRG | Recall@50: 43.8594 -> 62.4389 | Helped |
| increase shots from 1 to 5 for Frozen-in-Time | Frozen-in-Time · MOMA-LRG few-shot sub-activity | accuracy: 26.50 -> 26.30 | No clear effect |
Closest papers by meaning
- MOMA: Multi-Object Multi-Actor Activity Parsing (NeurIPS 2021 · similarity 0.92)
- LLaVAction: evaluating and training multi-modal large language models for action understanding (ICLR 2026 · similarity 0.87)
- FineGym: A Hierarchical Video Dataset for Fine-Grained Action Understanding (CVPR 2020 · similarity 0.87)
- Ego4D Goal-Step: Toward Hierarchical Understanding of Procedural Activities (NeurIPS 2023 · similarity 0.87)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.86)
- FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment (CVPR 2024 · similarity 0.85)
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.