Research map / Detection and segmentation / Open-set and few-shot detection
SEAL: Self-supervised Embodied Active Learning using Exploration and 3D Consistency
Devendra Singh Chaplot, Murtaza Dalal, Saurabh Gupta, Jitendra Malik, Ruslan Salakhutdinov
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 |
|---|---|---|---|
| replace random policy with Gainful Curiosity exploration policy (ablation) | SEAL · Gibson test Generalization | object detection AP50: 35.43 -> 40.02 | Improved |
| replace frontier exploration with Gainful Curiosity (ablation) | SEAL · Gibson test Generalization | object detection AP50: 37.39 -> 40.02 | Improved |
| replace Active Neural SLAM with Gainful Curiosity (ablation) | SEAL · Gibson test Generalization | object detection AP50: 38.90 -> 40.02 | Improved |
| replace Semantic Curiosity with Gainful Curiosity (ablation) | SEAL · Gibson test Specialization | object detection AP50: 39.21 -> 41.23 | Improved |
| replace self-training with 3DLabelProp 3D-consistent labeling (ablation) | SEAL · Gibson test Generalization | object detection AP50: 35.35 -> 40.02 | Improved |
| replace optical flow label propagation with 3DLabelProp (ablation) | SEAL · Gibson test Specialization | object detection AP50: 35.92 -> 41.23 | Improved |
| explore each test environment for one extra episode (Specialization) | SEAL · Gibson test | object detection AP50: 40.02 -> 41.23 | Improved |
| swap pretrained Mask-RCNN for SEAL-improved perception in SemExp | SemExp object goal navigation · Gibson Object Goal Navigation | success rate: 0.544 -> 0.627 | Improved |
| self-training or optical-flow baselines with other exploration policies | Mask-RCNN · Gibson test Generalization | object detection AP50: 34.82 -> 33.41 to 35.85 | No clear effect |
| fine-tune with k=10 human-labeled frames without label propagation | Mask-RCNN · Gibson test | AP50 | No clear effect |
Later papers that compared with this method
3 later accepted papers report Seal as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- LargeAD: Large-Scale Cross-Sensor Data Pretraining for Autonomous Driving (TPAMI 2025 · better on 1 of 1 shared result)
- Building a Strong Pre-Training Baseline for Universal 3D Large-Scale Perception (CVPR 2024 · better on 7 of 7 shared results)
- Three Pillars Improving Vision Foundation Model Distillation for Lidar (CVPR 2024 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EmbodiedScan: A Holistic Multi-Modal 3D Perception Suite Towards Embodied AI (CVPR 2024 · similarity 0.86)
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · similarity 0.85)
- Grounding Video Models to Actions through Goal Conditioned Exploration (ICLR 2025 · similarity 0.85)
- Online Segment Any 3D Thing as Instance Tracking (NeurIPS 2025 · similarity 0.85)
- Self-Improving Embodied Foundation Models (NeurIPS 2025 · similarity 0.84)
- Pre-trained Text-to-Image Diffusion Models Are Versatile Representation Learners for Control (NeurIPS 2024 · similarity 0.84)
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.
