Regarding the evaluation of reconstruction quality

[Sierkinhane]

Hi, thank you for the nice work. :)

I'd like to know whether you have evaluated the reconstruction quality on the commonly used benchmarks like ImageNet and UCF. And how about the reconstruction quality of the small faces and texts.

[wilson1yan]

We evaluate our model on ImageNet in the paper, but not on UCF. UCF in the code is primarily there as an example if you want to train your own model.

If I remember correctly, the pretrained models can do smaller faces / texts to some extent, but still struggle if they're too small. The pretrained models are also available to use if you want to try on your own data of interest