search

DeepXuan / Dn-Dp

47 stars 5 forks source link

咨询关于sr_training中部分的问题 #3

Open jjjjjgggj opened 8 months ago

jjjjjgggj commented 8 months ago

sr_img = Metrics.tensor2img(visuals['SR'], out_type=out_type) # uint8 hr_img = Metrics.tensor2img(visuals['HR'], out_type=out_type) # uint8 lr_img = Metrics.tensor2img(visuals['LR'], out_type=out_type) # uint8 fake_img = Metrics.tensor2img(visuals['INF'], out_type=out_type) # uint8

作者你好，我有问题想咨询一下，我按照你的要求进行训练，我没看懂这个sr_img和fake_img代表什么意思。然后代码在每10000个iter进行一次推断，然后我查看结果时，sr_img是一张全黑的图片

DeepXuan commented 8 months ago

fake_img: upsamled lr image by bicubic interpolation sr_img: inference results

Please check if the ‘save_img’ function is suitable for your data. It's also possible that the training crashed, resulting in the generation of an all-zero image.