magvit2_3d_train

[allenzg]

Thank you for your open-source work. I encountered some issues while implementing your code and would greatly appreciate your guidance on the following points:

In the code snippet:

weights = torch.load(ckpt_path, map_location="cpu")

model = resnet50()

model = vgg16()

model.load_state_dict(weights)

I noticed that the naming in the config file is related to ResNet, but the actual code appears to involve VGG16 weights. Could you please clarify this inconsistency?

Regarding inflate_ckpt_path and pretrained in the magvit2_3d_train_config, could you kindly explain the appropriate parameter settings and file names required?

[Jason3900]

I found that the performance varies, you can try both to find out which one is better.