Research map / Vision-language and multimodal / Human and embodied modeling
PaStaNet: Toward Human Activity Knowledge Engine
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 PaSta-based linear reasoning to Pairwise | HICO image-based activity recognition · HICO | mAP: 39.9 -> 46.3 | Helped |
| Add PaSta-based linear reasoning to TIN | HICO-DET · HICO-DET Default Full | mAP: 17.03 -> 22.65 | Helped |
| Pre-train with PaStaNet and add PaStaNet-Linear | TIN · V-COCO Scenario 1 | AP_role: 47.8 -> 51.0 | Helped |
| Pre-train with PaStaNet and add PaStaNet-Linear | LFB-Res-101-baseline · Image-based AVA | mAP: 23.3 -> 24.3 | Helped |
| Use ground-truth PaSta probabilities rather than predicted probabilities (ablation) | TIN+PaStaNet*-Linear · HICO-DET Default Full | mAP: 22.12 -> 34.86 | Helped |
| Replace linear reasoning with MLP reasoning (ablation) | TIN+PaStaNet* · HICO-DET Default Full | mAP: 22.12 -> 21.59 | Hurt |
| Replace linear reasoning with GCN reasoning (ablation) | TIN+PaStaNet* · HICO-DET Default Full | mAP: 22.12 -> 21.73 | Hurt |
| Replace linear reasoning with sequential LSTM reasoning (ablation) | TIN+PaStaNet* · HICO-DET Default Full | mAP: 22.12 -> 21.64 | Hurt |
| Replace linear reasoning with tree-structured passing (ablation) | TIN+PaStaNet* · HICO-DET Default Full | mAP: 22.12 -> 21.36 | Hurt |
| Remove part attention (ablation) | PaStaNet*-Linear on HICO-DET · HICO-DET | mAP: 0.21 mAP | Hurt |
| Replace BERT language features with Gaussian noise (ablation) | PaStaNet*-Linear on HICO-DET · HICO-DET | mAP: 20.80 mAP | Hurt |
| Replace BERT language features with Word2Vec (ablation) | PaStaNet*-Linear on HICO-DET · HICO-DET | mAP: 21.95 mAP | Hurt |
Closest papers by meaning
- Symbol-LLM: Leverage Language Models for Symbolic System in Visual Human Activity Reasoning (NeurIPS 2023 · similarity 0.84)
- MOMA-LRG: Language-Refined Graphs for Multi-Object Multi-Actor Activity Parsing (NeurIPS 2022 · similarity 0.83)
- MOMA: Multi-Object Multi-Actor Activity Parsing (NeurIPS 2021 · similarity 0.83)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.81)
- From Isolated Islands to Pangea: Unifying Semantic Space for Human Action Understanding (CVPR 2024 · similarity 0.81)
- Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition (NeurIPS 2025 · similarity 0.80)
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.