Research map / Detection and segmentation / Open-set and few-shot detection
Tuning Computer Vision Models With Task Rewards
André Susano Pinto, Alexander Kolesnikov, Yuge Shi, Lucas Beyer, Xiaohua Zhai
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 |
|---|---|---|---|
| Tune the pretrained model with a PQ-related reward | UViM panoptic segmentation · COCO panoptic val | PQ (%): 43.1 → 46.1 | Improved |
| Tune the pretrained detector with an mAP-related reward | ViT-B/16 detection MLE model · COCO | mAP (%): 39.2 → 54.3 | Improved |
| Tune the pretrained detector with an mAP-related reward | ViT-B/16 detection MLE model · COCO | AR@100 (%): 54.4 → 67.2 | Improved |
| Tune the pretrained detector with a recall reward | ViT-B/16 detection MLE model · COCO | AR@100 (%): 54.4 → 68.4 | Improved |
| Tune the colorization model with a vivid-color reward | UViM colorization · ImageNet colorization | Vivid-color reward term: 0.46 to 0.97 | Improved |
| Tune the colorization model with a color-diversity reward | UViM colorization · ImageNet colorization | Hue entropy: 1.03 to 1.84 | Improved |
| Tune the captioning model with CIDEr reward | Ours (ViT-B) · COCO caption test split | CIDEr: 120.0 → 134.5 | Improved |
| Tune the captioning model with CIDEr reward | Ours (ViT-L) · COCO caption test split | CIDEr: 121.7 → 138.7 | Improved |
| Use reward tuning rather than selecting the highest-likelihood sample from many draws | COCO captioning · COCO caption validation | reward: 10000 | Improved |
| Continue mAP reward tuning beyond the initial 20k steps | ViT-B/16 detection MLE model · COCO object detection | mAP (%): 40.2% to 52.3%; 52.7% at 40k steps; 53.2% at 60k steps | Improved |
Closest papers by meaning
- TTRV: Test-Time Reinforcement Learning for Vision Language Models (CVPR 2026 · similarity 0.87)
- Fine-Grained Verifiers: Preference Modeling as Next-token Prediction in Vision-Language Alignment (ICLR 2025 · similarity 0.87)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.86)
- InstructCV: Instruction-Tuned Text-to-Image Diffusion Models as Vision Generalists (ICLR 2024 · similarity 0.86)
- BalCapRL : A Balanced Framework for RL-Based MLLM Image Captioning (NeurIPS 2026 · similarity 0.86)
- Confidence-aware Reward Optimization for Fine-tuning Text-to-Image Models (ICLR 2024 · similarity 0.86)
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.
