Evaluation Metrics

Frechet Inception Distance (FID) is commonly used to evaluate generative model. It employs an Inception Model (pretrained on ImageNet) to extract features from both real and synthesized images.

Inception Model

For PGGAN, StyleGAN, etc, they use inception model from the TensorFlow Models repository, whose implementation is slightly different from that of torchvision. Hence, to make the evaluation metric comparable between different training frameworks (i.e., PyTorch and TensorFlow), we modify torchvision/models/inception.py as inception.py. The ported pre-trained weight is borrowed from this repo.

NOTE: We also support using the model from torchvision to compute the FID. However, please be aware that the FID value from torchvision is usually ~1.5 smaller than that from the TensorFlow model.

Please use the following code to choose which model to use.

from metrics.inception import build_inception_model

inception_model_tf = build_inception_model(align_tf=True)
inception_model_pth = build_inception_model(align_tf=False)