Steps for Training VQ-VAE?

[aksg87]

Hello,

I wanted to confirm the steps for training a VQ-VAE on radiology data. Thank you for working on such an interesting and important application of VQ-VAE. Our research group is particularly interested in applications to oncologic imaging.

1. Sample data

• Is there sample imaging data in: https://surfdrive.surf.nl/files/index.php/s/xY7bwjrgfnhPCAt?path=%2F or would you suggest we use other radiology datasets?
• Also, if we use our own dataset do you have details on how these should be formatted or structured? (I.e. Dicom files or NIfTI files)

2. Training

• Would you recommend leaving the default parameters in the vqvae/train.py script (aside from the dataset)?
• Any suggestions here would be very much appreciated!

[aksg87]

Just an update!

I was able to get the train.py script running which is cool 🙂. Will share a few details in case anyone finds them helpful.

1. I created a folder with a bunch of .nrrd volume files which is the only input I am using right now for train.py

2. The code initially threw in an error as it expected a list for num_embeddings but the default is an int so I did the following. vqvae/model.py

       # assert len(args.num_embeddings) in (1, args.n_bottleneck_blocks)  # expects list     
       self.num_embeddings = [args.num_embeddings for _ in range(args.n_bottleneck_blocks)]
   
       # if len(args.num_embeddings) == 1:
       #     self.num_embeddings = [args.num_embeddings[0] for _ in range(args.n_bottleneck_blocks)]
       # else:
       #     self.num_embeddings = args.num_embeddings

3. I added a batch-size default of 1 as I otherwise received an error: vqvae/train.py

   parser.add_argument("--batch-size", type=int, default=1)

[27yw]

thanks for your answer! It helps a lot. I am trying to training it as well and could you help me with my dataset_path, pls? I wanna train the data named like this :BraTS2021_00377_t1.nii.gz. Does it mean that I put all the files in one dir and what do I need to change the code and where?