training for minst and other dataset

[huzhongyyuan]

Hello! Thanks for your excellent work. I'm wondering if the following command is correct for training OptVQ on the MNIST dataset，do there exist any related YAML documents?

Also, I'm curious about the difference between much optvq.yaml and vqbase.yaml，what scenarios are these files applied to
And if I want to train an OptVQ model with my own datasets, what steps should I follow?

Much thanks！！！！！！！！！！！！！！！！！

[zbr17]

Hello, for training with the MNIST dataset, you can refer to the two YAML files under configs/mnist:

• configs/mnist/optvq.yaml: This file uses the quantizer proposed by OptVQ, which employs optimal transport for quantization and ensures higher codebook utilization. We recommend using this one.
• configs/mnist/vqbase.yaml: This file serves as the baseline method for comparison with OptVQ, using nearest neighbor for quantization.

---

If you need to train on your own dataset, you can customize your own dataset loading module by referring to the structure of ImageNetDataset in src/optvq/data/dataset.py. If you encounter any further difficulties, please feel free to contact us.