Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Look-Ahead Before You Leap: End-to-End Active Recognition by Forecasting the Effect of Motion
Dinesh Jayaraman, Kristen Grauman
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 the look-ahead module to the active RNN | Active RNN · SUN360 (T=5) | accuracy (%): 65.32 -> 66.01 | Improved |
| add the look-ahead module to the active RNN | Active RNN · SUN360 (T=3) | accuracy (%): 57.52 -> 58.12 | Improved |
| add the look-ahead module to the active RNN | Active RNN · SUN360 (T=2) | accuracy (%): 50.76 -> 51.72 | Improved |
| add the look-ahead module to the active RNN | Active RNN · GERMS (T=3) | accuracy (%): 46.86 -> 47.99 | Improved |
| add the look-ahead module to the active RNN | Active RNN · GERMS (T=2) | accuracy (%): 47.30 -> 48.02 | Improved |
| learn motion policy with REINFORCE instead of random motions (ablation) | Random views (recurrent) · SUN360 (T=5) | accuracy (%): 55.64 -> 65.32 | Improved |
| learn motion policy with REINFORCE instead of random motions (ablation) | Random views (recurrent) · GERMS (T=3) | accuracy (%): 44.24 -> 46.86 | Improved |
| average per-timestep classifier outputs on top of the look-ahead RNN | Look-ahead active RNN · SUN360 (T=5) | accuracy (%): 66.01 -> 62.61 | Hurt |
| average per-timestep classifier outputs on top of the look-ahead RNN | Look-ahead active RNN · GERMS (T=3) | accuracy (%): 47.99 -> 48.31 | Improved |
| use recurrent aggregation instead of averaging over random views (ablation) | Random views (average) · SUN360 (T=5) | accuracy (%): 54.21 -> 55.64 | Improved |
| use recurrent aggregation instead of averaging over random views (ablation) | Random views (average) · GERMS (T=3) | accuracy (%): 46.97 -> 44.24 | Hurt |
| use multiple random views instead of a single view (ablation) | Single view (neural net) · SUN360 (T=5) | accuracy (%): 40.12 -> 54.21 | Improved |
Closest papers by meaning
- End-to-End Policy Learning for Active Visual Categorization (TPAMI 2018 · similarity 0.97)
- Active Vision for Early Recognition of Human Actions (CVPR 2020 · similarity 0.88)
- Breaking the Passive Learning Trap: An Active Perception Strategy for Human Motion Prediction (AAAI 2026 · similarity 0.88)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.86)
- Evidential Active Recognition: Intelligent and Prudent Open-World Embodied Perception (CVPR 2024 · similarity 0.85)
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · 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.
