get_all_resnet() function converts a list of PIL images from #7 to a list of ResNet-18 feature vectors utilizing the handy img2vec library.
A few things to note:
ResNet-18 expects 244x244 images, since it was developed for ImageNet, not the CIFAR dataset. Note the line self.scaler = transforms.Resize((244, 244)) in img2vec/img_to_vec.py. I do not know if scaling 32x32 images to 244x244 will cause any issues.
I don't have a CUDA-enabled machine at the moment. Pushing the commits now so that others can test them out.
get_all_resnet()
function converts a list of PIL images from #7 to a list of ResNet-18 feature vectors utilizing the handy img2vec library.A few things to note:
self.scaler = transforms.Resize((244, 244))
in img2vec/img_to_vec.py. I do not know if scaling 32x32 images to 244x244 will cause any issues.