Q-Future / Q-Align

③[ICML2024] [IQA, IAA, VQA] All-in-one Foundation Model for visual scoring. Can efficiently fine-tune to downstream datasets.
https://q-align.github.io
MIT License
195 stars 14 forks source link

aesthetic assessment is sensitive to the size of the input image #10

Closed Nastu-Ho closed 4 months ago

Nastu-Ho commented 4 months ago

image

teowu commented 4 months ago

Hi Natsu. This is due to the pre-processing mechanism of the model, that images with long edge <448 will be padded to 448, which affects their aesthetic quality.

Please try with larger inputs (which will be first padded to square and then resized to 448*448).