Research map / Reinforcement learning / Visual and representation RL
Semantic Curiosity for Active Visual Learning
Devendra Singh Chaplot, Helen Jiang, Saurabh Gupta, Abhinav Gupta
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 semantic curiosity exploration instead of random exploration to collect training data | detector fine-tuning · Matterport test (11 scenes), detector fine-tuned on sampled trajectories | AP50 average: 37.4 -> 39.96 | Improved |
| use semantic curiosity instead of Active Neural SLAM exploration | detector fine-tuning · Matterport test (11 scenes), detector fine-tuned on sampled trajectories | AP50 average: 38.5 -> 39.96 | Improved |
| use semantic curiosity instead of coverage exploration | detector fine-tuning · Matterport test (11 scenes), detector fine-tuned on sampled trajectories | AP50 average: 36.56 -> 39.96 | Improved |
| use semantic curiosity instead of object exploration reward | detector fine-tuning · Matterport test (11 scenes), detector fine-tuned on sampled trajectories | AP50 average: 35.26 -> 39.96 | Improved |
| use prediction-error curiosity instead of random exploration | detector fine-tuning · Matterport test (11 scenes), detector fine-tuned on sampled trajectories | AP50 average: 37.4 -> 37.26 | No clear effect |
| fine-tune detector on random-policy trajectories vs not fine-tuning | COCO-pretrained detector · Matterport test (11 scenes), detector fine-tuned on sampled trajectories | AP50 average: 31.72 -> 37.4 | Improved |
| use semantic curiosity instead of Active Neural SLAM for covering area | exploration policy · Replica trajectories (exploration analysis) | Explored area: 11.527 -> 9.726 | Hurt |
| use semantic curiosity instead of object exploration reward for detections count | exploration policy · Replica trajectories (exploration analysis) | Number of object detections: 382.27 -> 291.78 | Hurt |
| use semantic curiosity instead of Active Neural SLAM | exploration policy · Replica trajectories (exploration analysis) | Semantic curiosity reward: 3.589 -> 4.378 | Improved |
| use semantic curiosity instead of Coverage Exploration (pretrained detector mAP50 on sampled data) | exploration policy · Replica trajectories, pretrained detector (5 classes) | mAP50 of pretrained detector: 51.02 -> 39.2 | Mixed |
Closest papers by meaning
- Look Around and Learn: Self-Training Object Detection by Exploration (ECCV 2024 · similarity 0.90)
- Learn How to See: Collaborative Embodied Learning for Object Detection and Camera Adjusting (AAAI 2024 · similarity 0.88)
- Curiosity-driven Exploration by Self-supervised Prediction (ICML 2017 · similarity 0.87)
- Learning to Look Around: Intelligently Exploring Unseen Environments for Unknown Tasks (CVPR 2018 · similarity 0.87)
- OVOD-Agent: A Markov-Bandit Framework for Proactive Visual Reasoning and Self-Evolving Detection (CVPR 2026 · similarity 0.87)
- Selective Visual Representations Improve Convergence and Generalization for Embodied AI (ICLR 2024 · 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.
