AI Paper Atlas

Research map / Image generation / Image restoration

Quality-Aware Pre-Trained Models for Blind Image Quality Assessment

Kai Zhao, Kun Yuan, Ming Sun, Mading Li, Xing Wen

CVPR 2023 · Image generation › Image restoration · approach: Raw and low-light noise modeling · 6 citations (OpenAlex)

Read the paper

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
Use QPT pre-training instead of standard ImageNet-supervised pre-trainingResNet50 BIQA model · BIDSRCC: 0.8423 -> 0.8875Improved
Use QPT pre-training instead of standard ImageNet-supervised pre-trainingResNet50 BIQA model · CLIVESRCC: 0.8527 -> 0.8947Improved
Replace HyperIQA pre-trained weights with QPT weightsHyperIQA · BIDSRCC: 0.8665 -> 0.8982Improved
Replace TReS pre-trained weights with QPT weightsTReS · CLIVESRCC: 0.8436 -> 0.8666Improved
Increase QPT pre-training data from 20% to 100% of ImageNet (ablation)QPT ResNet50 · BIDSRCC: 0.8311 -> 0.8875Improved
Increase QPT pre-training data from 20% to 100% of ImageNet (ablation)QPT ResNet50 · CLIVESRCC: 0.8115 -> 0.8947Improved
Use ResNet-50 rather than ResNet-18 encoder (ablation)QPT · BIDSRCC: 0.8679 -> 0.8875Improved
Add inter-sample content-based negatives (ablation)QPT contrastive loss with intra-sample negatives · BIDSRCC: 0.8399 -> 0.8875Improved
Add intra-sample degradation-based negatives (ablation)QPT contrastive loss with inter-sample negatives · BIDSRCC: 0.8702 -> 0.8875Improved
Remove both inter- and intra-sample negative pairs (ablation)QPT contrastive loss · BIDSRCC: 0.8875 -> 0.7078Hurt
Add two-order degradation (ablation)QPT degradation space with skip and shuffle · BIDSRCC: 0.8828 -> 0.8875Improved
Add shuffle to degradation composition (ablation)QPT degradation space with skip · BIDSRCC: 0.8745 -> 0.8875Improved

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