Research map / Video / Video segmentation and tracking
AFF-ttention! Affordances and Attention models for Short-Term Object Interaction Anticipation
Lorenzo Mur-Labadia, Ruben Martinez-Cantin, Jose J Guerrero, 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 |
|---|---|---|---|
| add environment affordance (E.AFF) fusion (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP N+V: 10.75 -> 11.75 | Improved |
| add interaction hotspot confidence re-weighting (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP N: 21.71 -> 23.63 | Improved |
| add both affordances and hotspots (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 3.53 -> 3.77 | Improved |
| add both affordance modules (ablation) | StillFast · Ego4D-STA v1 val | Top-5 mAP All: 2.48 -> 2.95 | Improved |
| use EGO-TOPO trained NN classifier for affordance distribution instead of kNN zone matching (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 3.74 -> 2.64 | Hurt |
| replace mean temporal pooling with frame-guided temporal pooling attention (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 2.63 -> 3.39 | Improved |
| replace sum fusion with dual image-video attention (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 3.39 -> 3.47 | Improved |
| use one-directional cross-attention instead of dual cross-attention (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 3.53 -> 3.01 (I to V); 3.35 (V to I) | Hurt |
| extract per-frame DINOv2 video features with mean pooling instead of a video model (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 2.52 -> 2.19 | Hurt |
| fine-tune last blocks of DINOv2 image encoder (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP N: 20.08 -> 21.71 | Improved |
| use multi-head attention in temporal pooling and fusion (ablation) | STAformer · Ego4D-STA v1 val | Top-5 mAP All: 3.53 -> 3.85 | Improved |
| train on Ego4D v2 instead of v1 | STAformer · Ego4D-STA test | Top-5 mAP All: 4.06 -> 6.75 | Improved |
Closest papers by meaning
- Integrating Affordances and Attention Models for Short-Term Object Interaction Anticipation (TPAMI 2026 · similarity 0.98)
- Anticipating Object Interactions Via Aggregation and Distillation of Spatio-Temporal Knowledge From Vision Language Models (TPAMI 2026 · similarity 0.93)
- Summarize the Past to Predict the Future: Natural Language Descriptions of Context Boost Multimodal Object Interaction Anticipation (CVPR 2024 · similarity 0.90)
- FIction: 4D Future Interaction Prediction from Video (CVPR 2025 · similarity 0.87)
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.87)
- Gaze Beyond the Frame: Forecasting Egocentric 3D Visual Span (NeurIPS 2025 · 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.
