Reproduction of FID scores

[selin-yildirim]

Hello,

I have generated some images with MAR model by the code given at: demo/run_mar.ipynb. After generating a set of 8 images, my goal is to observe FID scores for these images. To do that, I borrowed the code here:

mar/engine_mar.py

Line 196 in fe470ac

metrics_dict = torch_fidelity.calculate_metrics(

When passing input2=None in the snippet below, I am receiving the value error below, because it tries to read input 2:

 input2 = None
            fid_statistics_file = 'fid_stats/adm_in256_stats.npz'
        else:
            raise NotImplementedError
        metrics_dict = torch_fidelity.calculate_metrics(
            input1=save_folder,
            input2=input2,
            fid_statistics_file=fid_statistics_file,
            cuda=True,
            isc=True,
            fid=True,
            kid=False,
            prc=False,
            verbose=False,
        )

ValueError: Input descriptor "input" field can be either an instance of Dataset, GenerativeModelBase class, or a string, such as a path to a name of a registered dataset (cifar10-train, ci
far10-val, cifar100-train, cifar100-val, stl10-train, stl10-test, stl10-unlabeled), a directory with file samples, or a path to an ONNX or PTH (JIT) module  

Could you please help us debug this issue? Thanks.

[tameemalaa]

Hello. Did you fix the issue?

[hu-jy-0406]

Hello, I think it's related to the version of torch_fidelity. Try to uninstall your current torch_fidelity and run pip install -e git+https://github.com/LTH14/torch-fidelity.git@master#egg=torch-fidelity. I've got the same problem and solved in this way.