AI Paper Atlas

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

ICML 2023 · Detection and segmentation › Open-set and few-shot detection · approach: Object-level representation learning · 13 citations (OpenAlex)

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 changedWhereMeasuredEffect
Tune the pretrained model with a PQ-related rewardUViM panoptic segmentation · COCO panoptic valPQ (%): 43.1 → 46.1Improved
Tune the pretrained detector with an mAP-related rewardViT-B/16 detection MLE model · COCOmAP (%): 39.2 → 54.3Improved
Tune the pretrained detector with an mAP-related rewardViT-B/16 detection MLE model · COCOAR@100 (%): 54.4 → 67.2Improved
Tune the pretrained detector with a recall rewardViT-B/16 detection MLE model · COCOAR@100 (%): 54.4 → 68.4Improved
Tune the colorization model with a vivid-color rewardUViM colorization · ImageNet colorizationVivid-color reward term: 0.46 to 0.97Improved
Tune the colorization model with a color-diversity rewardUViM colorization · ImageNet colorizationHue entropy: 1.03 to 1.84Improved
Tune the captioning model with CIDEr rewardOurs (ViT-B) · COCO caption test splitCIDEr: 120.0 → 134.5Improved
Tune the captioning model with CIDEr rewardOurs (ViT-L) · COCO caption test splitCIDEr: 121.7 → 138.7Improved
Use reward tuning rather than selecting the highest-likelihood sample from many drawsCOCO captioning · COCO caption validationreward: 10000Improved
Continue mAP reward tuning beyond the initial 20k stepsViT-B/16 detection MLE model · COCO object detectionmAP (%): 40.2% to 52.3%; 52.7% at 40k steps; 53.2% at 60k stepsImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy