Research map / Detection and segmentation / Weakly supervised detection and segmentation
Look Around and Learn: Self-Training Object Detection by Exploration
Gianluca Scarpellini, Stefano Rosa, Pietro Morerio, Lorenzo Natale, Alessio Del Bue
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 |
|---|---|---|---|
| use Look Around policy with Disagreement Reconciliation instead of off-the-shelf detector | Mask R-CNN · Gibson simulated | mAP@50: 40.33 -> 46.60 | Improved |
| use Disagreement Reconciliation perception instead of self-training with random goals | Random policy · Gibson simulated | mAP@50: 39.67 -> 41.88 | Improved |
| use plain self-training with the Look Around policy | Look Around · Gibson simulated | mAP@50: 40.33 -> 38.66 | Hurt |
| add instance-matching triplet loss (ablation) | Look Around + Disag. reconc. · Gibson simulated | mAP@50: 44.19 -> 46.60 | Improved |
| add soft distillation loss with alpha 0.7 (ablation) | Look Around + Disag. reconc. · Gibson simulated | mAP@50: 41.09 -> 46.60 | Improved |
| set distillation weight alpha to 1.0 instead of 0.7 (ablation) | Look Around + Disag. reconc. · Gibson simulated | mAP@50: 46.60 -> 41.45 | Hurt |
| add soft distillation loss with Semantic Curiosity policy (ablation) | Semantic Curiosity + Disag. reconc. · Gibson simulated | mAP@50: 43.16 -> 41.37 | Hurt |
| use entropy-based disagreement score instead of cosine, Euclidean or class-count scores (ablation) | Look Around policy reward · Gibson simulated | mAP@50: 45.65 / 42.44 / 40.58 -> 46.60 | Improved |
| use Look Around instead of SEAL pipeline | exploration policy · Gibson simulated | mAP@50: +3.59% over SEAL, +2.45% over Informative Trajectories | Improved |
| use SEAL perception instead of Disagreement Reconciliation with SEAL policy | SEAL policy · Gibson simulated | mAP@50: 43.01 vs 42.38; Inform 44.15 vs 43.70 | Mixed |
| deploy Look Around on a real robot with Disagreement Reconciliation | Mask R-CNN · Real robot test | mAP@50: 55.83 -> 65.62 | Improved |
| use human-guided goals instead of learned policy | real robot data collection · Real robot test | mAP@50: 65.62 -> 56.00 | Hurt |
Closest papers by meaning
- Semantic Curiosity for Active Visual Learning (ECCV 2020 · similarity 0.90)
- Learning to Look Around: Intelligently Exploring Unseen Environments for Unknown Tasks (CVPR 2018 · similarity 0.88)
- Learn How to See: Collaborative Embodied Learning for Object Detection and Camera Adjusting (AAAI 2024 · similarity 0.87)
- An Exploration of Embodied Visual Exploration (IJCV 2021 · similarity 0.87)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.85)
- OVOD-Agent: A Markov-Bandit Framework for Proactive Visual Reasoning and Self-Evolving Detection (CVPR 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.
