Research map / Video / Video-language understanding
Attend and Interact: Higher-Order Object Interactions for Video Understanding
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 object interactions with dot-product attentive selection (ablation) | SINet image-context model · Kinetics validation | Top-1 accuracy: 73.1 -> 74.2 | Helped |
| Model object inter-relationships using dot-product attention rather than alpha-attention (ablation) | SINet · Kinetics validation | Top-1 accuracy: 73.9 -> 74.2 | Helped |
| Replace image-feature LSTM aggregation with temporal SDP-Attention (ablation) | SINet image-context branch · Kinetics validation | Top-1 accuracy: 70.6 -> 71.1 | Helped |
| Add mean-pooled object features to image features (ablation) | SINet · Kinetics validation | Top-1 accuracy: 71.1 -> 73.1 | Helped |
| Use object triplets instead of individual-object mean-pooling (ablation) | Kinetics object interaction model · Kinetics | Top-1 accuracy: 73.1 -> 72.9 | Hurt |
| Use object pairs instead of individual-object mean-pooling (ablation) | Kinetics object interaction model · Kinetics | FLOP: ~1.9 -> ~18.3 | Hurt |
| Use object triplets instead of individual-object mean-pooling (ablation) | Kinetics object interaction model · Kinetics | FLOP: ~1.9 -> ~77.0 | Hurt |
| Increase attentive selection groups from K=1 to K=2 (ablation) | SINet · Kinetics | Top-1 accuracy: 73.9 -> 74.2 | Helped |
| Increase attentive selection groups from K=2 to K=3 (ablation) | SINet · Kinetics | Top-5 accuracy: 91.5 -> 91.7 | Helped |
| Add object interaction features to image features (ablation) | SINet-Caption · ActivityNet Captions validation | CIDEr-D: 43.12 -> 44.81 | Helped |
| Add co-attention over object interactions (ablation) | SINet-Caption with image and object features · ActivityNet Captions validation (average 1st and 2nd) | CIDEr-D: 44.81 -> 44.84 | No clear effect |
| Replace C3D features with ResNeXt features (ablation) | SINet-Caption image-only model · ActivityNet Captions validation | CIDEr-D: 38.22 -> 43.12 | Helped |
Closest papers by meaning
- SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos (CVPR 2024 · similarity 0.86)
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.86)
- HIG: Hierarchical Interlacement Graph Approach to Scene Graph Generation in Video Understanding (CVPR 2024 · similarity 0.86)
- Open-Vocabulary Video Relation Extraction (AAAI 2024 · similarity 0.86)
- SAMA: Towards Multi-Turn Referential Grounded Video Chat with Large Language Models (NeurIPS 2025 · similarity 0.86)
- Capturing Closely Interacted Two-Person Motions with Reaction Priors (CVPR 2024 · 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.