add more configurations.

Author: ildoonet

README.md

@@ -31,25 +31,41 @@ transform_train.transforms.insert(0, RandAugment(N, M))
 
 ## Experiment
 
-We use same hyperparameters as the paper mentioned. 
+We use same hyperparameters as the paper mentioned. We observed similar results as reported. 
 
-### CIFAR-10
+You can run an experiment with, 
+
+```bash
+$ python RandAugment/train.py -c confs/wresnet28x10_cifar10_b256.yaml --save cifar10_wres28x10.pth
+```
+
+### CIFAR-10 Classification
 
 | Model             | Paper's Result | Ours         |
 |-------------------|---------------:|-------------:|
 | Wide-ResNet 28x10 | 97.3           | 97.4         |
-| Shake26 2x96d     | 
-| Pyramid272        | TODO           |
+| Shake26 2x96d     | 98.0           | 98.1         |
+| Pyramid272        | 98.5           |
 
-### CIFAR-100
+### CIFAR-100 Classification
 
 | Model             | Paper's Result | Ours         |
 |-------------------|---------------:|-------------:|
 | Wide-ResNet 28x10 | 83.3           | 83.3         |
 
-### ImageNet
+### SVHN Classification
 
-TODO
+| Model             | Paper's Result | Ours         |
+|-------------------|---------------:|-------------:|
+| Wide-ResNet 28x10 | 98.9           | TODO         |
+
+### ImageNet Classification
+
+| Model             | Paper's Result | Ours         |
+|-------------------|---------------:|-------------:|
+| ResNet-50         | 77.6 / 92.8    | TODO 
+| EfficientNet-B5   | 83.2 / 96.7    | TODO
+| EfficientNet-B7   | 84.4 / 97.1    | TODO
 
 ## References
 

RandAugment/data.py

@@ -5,6 +5,7 @@
 from PIL import Image
 
 from torch.utils.data import SubsetRandomSampler, Sampler
+from torch.utils.data.dataset import ConcatDataset
 from torchvision.transforms import transforms
 from sklearn.model_selection import StratifiedShuffleSplit
 from theconf import Config as C
@@ -80,6 +81,11 @@ def get_dataloaders(dataset, batch, dataroot, split=0.15, split_idx=0):
     elif dataset == 'cifar100':
         total_trainset = torchvision.datasets.CIFAR100(root=dataroot, train=True, download=True, transform=transform_train)
         testset = torchvision.datasets.CIFAR100(root=dataroot, train=False, download=True, transform=transform_test)
+    elif dataset == 'svhn':
+        trainset = torchvision.datasets.SVHN(root=dataroot, split='train', download=True, transform=transform_train)
+        extraset = torchvision.datasets.SVHN(root=dataroot, split='extra', download=True, transform=transform_train)
+        total_trainset = ConcatDataset([trainset, extraset])
+        testset = torchvision.datasets.SVHN(root=dataroot, split='test', download=True, transform=transform_test)
     elif dataset == 'imagenet':
         total_trainset = ImageNet(root=os.path.join(dataroot, 'imagenet-pytorch'), transform=transform_train)
         testset = ImageNet(root=os.path.join(dataroot, 'imagenet-pytorch'), split='val', transform=transform_test)

RandAugment/lr_scheduler.py

@@ -11,6 +11,8 @@ def adjust_learning_rate_resnet(optimizer):
 
     if C.get()['epoch'] == 90:
         return torch.optim.lr_scheduler.MultiStepLR(optimizer, [30, 60, 80])
+    elif C.get()['epoch'] == 180:
+        return torch.optim.lr_scheduler.MultiStepLR(optimizer, [60, 120, 160])
     elif C.get()['epoch'] == 270:   # autoaugment
         return torch.optim.lr_scheduler.MultiStepLR(optimizer, [90, 180, 240])
     else:

confs/pyramid272_cifar10_b64.yaml

@@ -0,0 +1,21 @@
+model:
+  type: pyramid
+  depth: 272
+  alpha: 200
+  bottleneck: True
+dataset: cifar10
+aug: randaugment
+randaug:
+  N: 3
+  M: 9
+
+cutout: 16
+batch: 64
+epoch: 1800
+lr: 0.05
+lr_schedule:
+  type: 'cosine'
+optimizer:
+  type: sgd
+  nesterov: True
+  decay: 0.00005

confs/resnet200_b256.yaml

@@ -0,0 +1,18 @@
+model:
+  type: resnet200
+dataset: imagenet
+aug: fa_reduced_imagenet
+cutout: 0
+batch: 256
+epoch: 270
+lr: 0.05
+lr_schedule:
+  type: 'resnet'
+  warmup:
+    multiplier: 2
+    epoch: 3
+optimizer:
+  type: sgd
+  nesterov: True
+  decay: 0.0001
+  clip: 0

confs/resnet50_b1024.yaml

@@ -0,0 +1,22 @@
+model:
+  type: resnet50
+dataset: imagenet
+aug: randaugment
+randaug:
+  N: 2
+  M: 9
+
+cutout: 0
+batch: 1024
+epoch: 180  # 270
+lr: 0.1
+lr_schedule:
+  type: 'resnet'
+  warmup:
+    multiplier: 4
+    epoch: 3
+optimizer:
+  type: sgd
+  nesterov: True
+  decay: 0.0001
+  clip: 0

confs/shake26_2x96d_cifar10_b512.yaml

@@ -1,7 +1,11 @@
 model:
   type: shakeshake26_2x96d
 dataset: cifar10
-aug: fa_reduced_cifar10
+aug: randaugment
+randaug:
+  N: 3
+  M: 9
+
 cutout: 16
 batch: 512
 epoch: 1800

confs/wresnet28x10_svhn_b256.yaml

@@ -0,0 +1,21 @@
+model:
+  type: wresnet28_10
+dataset: svhn
+aug: randaugment
+randaug:
+  N: 3
+  M: 7    # from appendix
+
+cutout: 16
+batch: 256
+epoch: 160
+lr: 0.005
+lr_schedule:
+  type: 'cosine'
+  warmup:
+    multiplier: 2
+    epoch: 5
+optimizer:
+  type: sgd
+  nesterov: True
+  decay: 0.001