Hyper-parameters for reproducing the results on ImageNet about repdistiller HOT 9 OPEN

The validation curve is attached and the best acc is 70.434.
Hope for your reply.

from repdistiller.

This is a great work. However, when I try to reproduce results on the ImageNet dataset, there is a 1% accuracy gap between mine and that in your paper.

Would you mind providing the hyper-parameters for training on ImageNet?

Here is mine:
-r 1.0
-a 0.0
-b 0.8
--trial 1
--weight_decay 0.0001
--learning_rate 0.1
--epochs 100
--lr_decay_epochs 30,60,90
--print_freq 500
--batch_size 256 \

eight 1080Ti GPUs are used.
Thanks!

I met the problem today, have you solve this problem? I think the dataset is wrong, i will try to rewrite the code

from repdistiller.

This is a great work. However, when I try to reproduce results on the ImageNet dataset, there is a 1% accuracy gap between mine and that in your paper.
Would you mind providing the hyper-parameters for training on ImageNet?
Here is mine:
-r 1.0
-a 0.0
-b 0.8
--trial 1
--weight_decay 0.0001
--learning_rate 0.1
--epochs 100
--lr_decay_epochs 30,60,90
--print_freq 500
--batch_size 256
eight 1080Ti GPUs are used.
Thanks!

I met the problem today, have you solve this problem? I think the dataset is wrong, i will try to rewrite the code

Not solved yet. I gave up.
Similar results were obtained in this repo.
https://github.com/yoshitomo-matsubara/torchdistill/tree/master/configs/official/ilsvrc2012/yoshitomo-matsubara/rrpr2020

from repdistiller.

This is a great work. However, when I try to reproduce results on the ImageNet dataset, there is a 1% accuracy gap between mine and that in your paper.
Would you mind providing the hyper-parameters for training on ImageNet?
Here is mine:
-r 1.0
-a 0.0
-b 0.8
--trial 1
--weight_decay 0.0001
--learning_rate 0.1
--epochs 100
--lr_decay_epochs 30,60,90
--print_freq 500
--batch_size 256
eight 1080Ti GPUs are used.
Thanks!

I met the problem today, have you solve this problem? I think the dataset is wrong, i will try to rewrite the code

Not solved yet. I gave up.
Similar results were obtained in this repo.
https://github.com/yoshitomo-matsubara/torchdistill/tree/master/configs/official/ilsvrc2012/yoshitomo-matsubara/rrpr2020

i found test_set = datasets.ImageFolder(test_folder, transform=test_transform) is wrong, test_set gets incorrect labels, i guess something has been done or processed in author's work. So i am coding this part, not difficult, expecting a normal result. ^v^

from repdistiller.

class TestImageDataset(Dataset):
    def __init__(self, root, transform, classes2label=None):
        super(TestImageDataset, self).__init__()
        self.root = root
        self.transform = transform
        self.classes2label = classes2label
        self.image_file_list, self.label_list = self.parse_txt()

    def __len__(self):
        return len(self.image_file_list)

    def __getitem__(self, idx):
        file = os.path.join(self.root,'images', self.image_file_list[idx])
        img = Image.open(file).convert('RGB')
        label = torch.tensor(int(self.label_list[idx])).long()
        if self.transform is not None:
            img = self.transform(img)
        return img, label

    def parse_txt(self):
        annotation_path = os.path.join(self.root, 'val_annotations.txt')
        image_file_list = []
        label_list = []

        with open(annotation_path, 'r') as f:
            contents = f.readlines()
        for content in contents:
            image_file, classes_name = content.split('\t')[:2]
            image_file_list.append(image_file)
            label = self.classes2label[classes_name]
            label_list.append(label)
        return image_file_list, label_list

if is_instance:
    train_set = ImageFolderInstance(train_folder, transform=train_transform)
    n_data = len(train_set)

else:
    train_set = datasets.ImageFolder(train_folder, transform=train_transform)
test_set = TestImageDataset(root=test_folder, transform=test_transform, classes2label=train_set.class_to_idx)

@kumamonatseu This code works! ^v^

from repdistiller.

@shaoeric thanks for your reply. it seems that codes of the dataset are already rewritten in another repo. anyway, hope for good news from you!

from repdistiller.

class TestImageDataset(Dataset):
    def __init__(self, root, transform, classes2label=None):
        super(TestImageDataset, self).__init__()
        self.root = root
        self.transform = transform
        self.classes2label = classes2label
        self.image_file_list, self.label_list = self.parse_txt()

    def __len__(self):
        return len(self.image_file_list)

    def __getitem__(self, idx):
        file = os.path.join(self.root,'images', self.image_file_list[idx])
        img = Image.open(file).convert('RGB')
        label = torch.tensor(int(self.label_list[idx])).long()
        if self.transform is not None:
            img = self.transform(img)
        return img, label

    def parse_txt(self):
        annotation_path = os.path.join(self.root, 'val_annotations.txt')
        image_file_list = []
        label_list = []

        with open(annotation_path, 'r') as f:
            contents = f.readlines()
        for content in contents:
            image_file, classes_name = content.split('\t')[:2]
            image_file_list.append(image_file)
            label = self.classes2label[classes_name]
            label_list.append(label)
        return image_file_list, label_list

if is_instance:
    train_set = ImageFolderInstance(train_folder, transform=train_transform)
    n_data = len(train_set)

else:
    train_set = datasets.ImageFolder(train_folder, transform=train_transform)
test_set = TestImageDataset(root=test_folder, transform=test_transform, classes2label=train_set.class_to_idx)

@kumamonatseu This code works! ^v^

May I ask what the 'val_annotations.txt' is? I can't find it in ilsvrc2012. Thanks!

from repdistiller.

class TestImageDataset(Dataset):
    def __init__(self, root, transform, classes2label=None):
        super(TestImageDataset, self).__init__()
        self.root = root
        self.transform = transform
        self.classes2label = classes2label
        self.image_file_list, self.label_list = self.parse_txt()

    def __len__(self):
        return len(self.image_file_list)

    def __getitem__(self, idx):
        file = os.path.join(self.root,'images', self.image_file_list[idx])
        img = Image.open(file).convert('RGB')
        label = torch.tensor(int(self.label_list[idx])).long()
        if self.transform is not None:
            img = self.transform(img)
        return img, label

    def parse_txt(self):
        annotation_path = os.path.join(self.root, 'val_annotations.txt')
        image_file_list = []
        label_list = []

        with open(annotation_path, 'r') as f:
            contents = f.readlines()
        for content in contents:
            image_file, classes_name = content.split('\t')[:2]
            image_file_list.append(image_file)
            label = self.classes2label[classes_name]
            label_list.append(label)
        return image_file_list, label_list

if is_instance:
    train_set = ImageFolderInstance(train_folder, transform=train_transform)
    n_data = len(train_set)

else:
    train_set = datasets.ImageFolder(train_folder, transform=train_transform)
test_set = TestImageDataset(root=test_folder, transform=test_transform, classes2label=train_set.class_to_idx)

@kumamonatseu This code works! ^v^

May I ask what the 'val_annotations.txt' is? I can't find it in ilsvrc2012. Thanks!

very glad to help you, but i run the code just on tiny-imagenet, so sorry

from repdistiller.

class TestImageDataset(Dataset):
    def __init__(self, root, transform, classes2label=None):
        super(TestImageDataset, self).__init__()
        self.root = root
        self.transform = transform
        self.classes2label = classes2label
        self.image_file_list, self.label_list = self.parse_txt()

    def __len__(self):
        return len(self.image_file_list)

    def __getitem__(self, idx):
        file = os.path.join(self.root,'images', self.image_file_list[idx])
        img = Image.open(file).convert('RGB')
        label = torch.tensor(int(self.label_list[idx])).long()
        if self.transform is not None:
            img = self.transform(img)
        return img, label

    def parse_txt(self):
        annotation_path = os.path.join(self.root, 'val_annotations.txt')
        image_file_list = []
        label_list = []

        with open(annotation_path, 'r') as f:
            contents = f.readlines()
        for content in contents:
            image_file, classes_name = content.split('\t')[:2]
            image_file_list.append(image_file)
            label = self.classes2label[classes_name]
            label_list.append(label)
        return image_file_list, label_list

if is_instance:
    train_set = ImageFolderInstance(train_folder, transform=train_transform)
    n_data = len(train_set)

else:
    train_set = datasets.ImageFolder(train_folder, transform=train_transform)
test_set = TestImageDataset(root=test_folder, transform=test_transform, classes2label=train_set.class_to_idx)

@kumamonatseu This code works! ^v^

May I ask what the 'val_annotations.txt' is? I can't find it in ilsvrc2012. Thanks!

very glad to help you, but i run the code just on tiny-imagenet, so sorry

Thanks for reply

from repdistiller.