search

TencentARC / VQFR

ECCV 2022, Oral, VQFR: Blind Face Restoration with Vector-Quantized Dictionary and Parallel Decoder
https://ycgu.site/projects/vqfr
Other
322 stars 40 forks source link

questions about the evalutions of VQGAN. #15

Open zwb0 opened 1 year ago

zwb0 commented 1 year ago

Hi, Could I ask about the evaluation of the results of VQGAN. I generate the results by, python -u vqfr/test.py -opt options/test/VQGAN/test_vqgan_v1.yml

Then I evaluated the metrics by the commands as the figure below, but the results are lower than the results of VQFR provided in the paper. Since the VQGAN is an autoencoder of the HR images, is it correct that the results of VQGAN are lower than VQFR? Or do you know where did I evaluated wrong? Thanks in advance! 结果

zwb0 commented 1 year ago

Sorry, I found the mistake, and my evaluation of the VQGAN is shown below. Am I correct since there are some metrics are still worse than stage 2? Thanks! 结果

helloful commented 1 year ago

你这个把第一阶段的指标计算出来有什么意义吗？第一阶段就是一个VAE，输入高质量的，重建高质量的，你这个第一阶段就是输入低质量，重建高质量吗？