search

Justin-Tan / high-fidelity-generative-compression

Pytorch implementation of High-Fidelity Generative Image Compression + Routines for neural image compression
Apache License 2.0
411 stars 77 forks source link

About the model choice #25

Open JXH-SHU opened 3 years ago

JXH-SHU commented 3 years ago

Firstly, thank you for offering this good code project. When I try to test the pre_trained model on my own dataset, I find the hific_low.pt can compress images at a level of bpp0.1441, hific_med.pt at a level of bpp0.2865, and hific_hi.pt at a level of bpp0.3041. The hific_med and hific_hi offer a nearly same compression rate, through the image quality is different.

Then when I train the model on my own dataset, could you share some experience with me for how to choose model trained for different epoches.

JXH-SHU commented 3 years ago

Because the quality and compression rate are fluctuated.