Research map / Reinforcement learning / Visual and representation RL
Learning Transferable Reward for Query Object Localization with Policy Adaptation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Adapt the policy with the query examples and unlabeled test images instead of fine-tuning on the small labeled exemplar set | Localization agent · Corrupted MNIST, new digits | CorLoc: 83.7 -> 97.4 | Helped |
| Adapt the policy using the query examples and unlabeled target images | Localization agent · Corrupted MNIST, new backgrounds | CorLoc: 56.4 -> 99.3 | Helped |
| Adapt the policy using the query examples and unlabeled target images | Localization agent · COCO, novel classes | CorLoc: 22.3 -> 39.3 | Helped |
| Use the ordinal reward localization method with test-time adaptation rather than TFA w/ fc | COCO few-shot localization · COCO, novel classes | CorLoc: 14.4 -> 39.3 | Helped |
| Add ordinal pre-training to the Faster RCNN embedding (ablation) | RL localization agent with Faster RCNN embedding · COCO, source classes | CorLoc: 25.93 -> 66.67; 13.17 -> 72.46; 12.16 -> 61.59; 16.18 -> 60.29; 28.57 -> 71.25 | Helped |
| Add ordinal structure to the embedding (ablation) | Selective-search ranking localization · COCO | CorLoc | Helped |
| Replace RL localization with selective-search ranking by embedding distance (ablation) | Ordinal-reward localization · COCO | CorLoc | Hurt |
| Use a continuous unsigned embedding reward instead of applying the sign operation (ablation) | AE+Ord+Embed localization agent · Corrupted MNIST, digit 4 | CorLoc: 90.92\% -> 94.36\% | Helped |
| Use a continuous unsigned embedding reward instead of applying the sign operation (ablation) | AE+Ord+Embed localization agent · Corrupted MNIST, other digits | CorLoc: 82.43\% -> 88.64\% | Helped |
| Use policy gradient with an RNN instead of DQN (ablation) | Localization agent · Corrupted MNIST, digit 4 | CorLoc: $88.80_{1.6}$ -> $94.68_{0.9}$ | Helped |
| Use policy gradient with an RNN instead of DQN (ablation) | Localization agent · Corrupted MNIST, other digits | CorLoc: $84.21_{2.0}$ -> $89.05_{1.7}$ | Helped |
| Use the learned embedding distance reward rather than IoU reward (ablation) | AE+Ord localization agent · Corrupted MNIST | CorLoc | Helped |
Closest papers by meaning
- RELO: Reinforcement Learning to Localize for Visual Object Tracking (ICML 2026 · similarity 0.88)
- Tree-Structured Reinforcement Learning for Sequential Object Localization (NeurIPS 2016 · similarity 0.87)
- Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction (ICLR 2025 · similarity 0.84)
- Test-Time Adaptation for Online Vision-Language Navigation with Feedback-based Reinforcement Learning (ICML 2025 · similarity 0.83)
- TTRV: Test-Time Reinforcement Learning for Vision Language Models (CVPR 2026 · similarity 0.83)
- RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations (CVPR 2025 · similarity 0.83)
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.