DC-Bench includes most of the commonly used models to evaluate the performance of synthetic datasets.

But it's also easy to extend to add new models.

Step 1: Create the new model

You can create the new model in whatever way you like as long as it's a subclass of the PyTorch nn.Module

Step 2: Integrate

After you are done creating the model, put it under the networks folder. Then open network_utils.py in the networks folder and add your model similar to the following code

elif model_name == 'resnet18':
  return ResNet18(channel=channel, num_classes=num_classes)

Step 3: Run the eval

Now you have integrated your new model, you can starting using the new model as

python evaluator/evaluator.py --method DC  --model NEW_MODEL