search

xw-hu / DSC

CVPR 2018 (Oral) & TPAMI 2019
141 stars 35 forks source link

Shadow removal results without the pre-trained filters #8

Closed hieulem closed 5 years ago

hieulem commented 5 years ago

Hi, it's a not an issue. I am just curious about the shadow removal results when your networks are trained from scratch.

xw-hu commented 5 years ago

We use the VGG network trained on ImageNet to initial the parameters.

hieulem commented 5 years ago

Hi, thank you for the prompt reply. I understand that from the paper as well. I am just curious about its impact, i.e., what is the performance of the model if it doesn't get initialized from the VGG pre-trained parameters.

xw-hu commented 5 years ago

You can have a try. I think it needs more training iterations.