Evaluate on custom dataset

[daniiki]

Hi, how can you evaluate different models on a custom dataset?

[mathildecaron31]

Hi,
You will need to adapt this class to your own dataset:

swav/src/multicropdataset.py

Lines 19 to 62 in 82bddbb

	class MultiCropDataset(datasets.ImageFolder):
	def __init__(
	self,
	data_path,
	size_crops,
	nmb_crops,
	min_scale_crops,
	max_scale_crops,
	size_dataset=-1,
	return_index=False,
	):
	super(MultiCropDataset, self).__init__(data_path)
	assert len(size_crops) == len(nmb_crops)
	assert len(min_scale_crops) == len(nmb_crops)
	assert len(max_scale_crops) == len(nmb_crops)
	if size_dataset >= 0:
	self.samples = self.samples[:size_dataset]
	self.return_index = return_index
	trans = []
	color_transform = transforms.Compose([get_color_distortion(), RandomGaussianBlur()])
	mean = [0.485, 0.456, 0.406]
	std = [0.228, 0.224, 0.225]
	for i in range(len(size_crops)):
	randomresizedcrop = transforms.RandomResizedCrop(
	size_crops[i],
	scale=(min_scale_crops[i], max_scale_crops[i]),
	)
	trans.extend([transforms.Compose([
	randomresizedcrop,
	transforms.RandomHorizontalFlip(p=0.5),
	color_transform,
	transforms.ToTensor(),
	transforms.Normalize(mean=mean, std=std)])
	] * nmb_crops[i])
	self.trans = trans
	def __getitem__(self, index):
	path, _ = self.samples[index]
	image = self.loader(path)
	multi_crops = list(map(lambda trans: trans(image), self.trans))
	if self.return_index:
	return index, multi_crops
	return multi_crops

[daniiki]

Hi again,
thanks for answering! Is there a way to use a pretrained model on deepclusteringv2 and evaluate on your own dataset. I want to test how good the clustering results are on my custom dataset without finetuning/training?
I know that you provide two deepclusteringv2 pretrained models, but which file to use for testing?

[mathildecaron31]

I think you should use the best deepcluster-v2 model, that is https://dl.fbaipublicfiles.com/deepcluster/deepclusterv2_800ep_pretrain.pth.tar

[mathildecaron31]

Closing for no activity - feel free to reopen the issue if you still need assistance.