Closed xunmeibuyue closed 2 years ago
Hi, I think the issue may be caused by some broken images in your dataset. Thus I recommend checking your target dataset first.
Thanks, perhaps because I used the camera JPEG outputs as rendered sRGB image.
Will feedback soon.
You are right. I find it is caused by wrong flip of some images. If training model using other datasets, the flip operation in this line:https://github.com/yzxing87/Invertible-ISP/blob/344dd333dd2a075f6a9e4ffc445dc387ca3014c4/data/data_preprocess.py#L53 should closely follow the operation of getting raw sensor data, i.e. this line:https://github.com/yzxing87/Invertible-ISP/blob/344dd333dd2a075f6a9e4ffc445dc387ca3014c4/data/data_preprocess.py#L49
Anyway, Thanks!
I'm trying to train the model on another dataset. But I have encountered the following problem:
I have searched for this problem on Google and stackoverflow, but the answers only mention that it maybe the wrong output dimension of certain layers.
So are there any fixed parameters of image size in this code? Would you mind having a look at it and pointing out the problem? Thanks!