Hi @knazeri
I have collected new images and downloaded mask images which you've used.
I have some questions to start model training:
1). About image dataset, I am using JPEG images which downloaded from google's dataset:
I split 720 X 720 images from their 4K image, and trying to use it as image dataset.
Does it work?
2). I am not sure, but you've been using 3 channel images in testing. If there is any one image with 4 channels in train/test/val images, then it will cause any problem while training?
3). About model stage, I need to run 3 steps to get final model?
You've wrote in README:
EdgeConnect is trained in three stages: 1) training the edge model, 2) training the inpaint model and 3) training the joint model. To train the model:
It does mean that I need to run train.py for 3 times with stage 1,2,3 to get final model?
4). About checkpoints, I can't use your pretrained checkpoints since I am using new image dataset?
Looking forward to hearing from you as soon as possible.
Thank you
Hi @knazeri I have collected new images and downloaded mask images which you've used. I have some questions to start model training: 1). About image dataset, I am using JPEG images which downloaded from google's dataset: I split 720 X 720 images from their 4K image, and trying to use it as image dataset. Does it work?
2). I am not sure, but you've been using 3 channel images in testing. If there is any one image with 4 channels in train/test/val images, then it will cause any problem while training?
3). About model stage, I need to run 3 steps to get final model? You've wrote in README:
It does mean that I need to run
train.pyfor 3 times with stage 1,2,3 to get final model?4). About checkpoints, I can't use your pretrained checkpoints since I am using new image dataset?
Looking forward to hearing from you as soon as possible. Thank you