rename dataset_root to data_root

Author: ajbrock

make_hdf5.py

@@ -25,7 +25,7 @@ def prepare_parser():
     help='Which Dataset to train on, out of I128, I256, C10, C100;'
          'Append "_hdf5" to use the hdf5 version for ISLVRC (default: %(default)s)')
   parser.add_argument(
-    '--dataset_root', type=str, default='data',
+    '--data_root', type=str, default='data',
     help='Default location where data is stored (default: %(default)s)')
   parser.add_argument(
     '--batch_size', type=int, default=256,
@@ -58,7 +58,7 @@ def run(config):
   train_loader = utils.get_data_loaders(dataset=config['dataset'],
                                         batch_size=config['batch_size'],
                                         shuffle=False,
-                                        dataset_root=config['dataset_root'],
+                                        data_root=config['data_root'],
                                         use_multiepoch_sampler=False,
                                         **kwargs)[0]     
 
@@ -81,7 +81,7 @@ def run(config):
     y = y.numpy()
     # If we're on the first batch, prepare the hdf5
     if i==0:
-      with h5.File(config['dataset_root'] + '/ILSVRC%i.hdf5' % config['image_size'], 'w') as f:
+      with h5.File(config['data_root'] + '/ILSVRC%i.hdf5' % config['image_size'], 'w') as f:
         print('Producing dataset of len %d' % len(train_loader.dataset))
         imgs_dset = f.create_dataset('imgs', x.shape,dtype='uint8', maxshape=(len(train_loader.dataset), 3, config['image_size'], config['image_size']),
                                      chunks=(config['chunk_size'], 3, config['image_size'], config['image_size']), compression=config['compression']) 
@@ -92,7 +92,7 @@ def run(config):
         labels_dset[...] = y
     # Else append to the hdf5
     else:
-      with h5.File(config['dataset_root'] + '/ILSVRC%i.hdf5' % config['image_size'], 'a') as f:
+      with h5.File(config['data_root'] + '/ILSVRC%i.hdf5' % config['image_size'], 'a') as f:
         f['imgs'].resize(f['imgs'].shape[0] + x.shape[0], axis=0)
         f['imgs'][-x.shape[0]:] = x
         f['labels'].resize(f['labels'].shape[0] + y.shape[0], axis=0)

utils.py

@@ -248,7 +248,7 @@ def prepare_parser():
     help='Default location to store all weights, samples, data, and logs '
            ' (default: %(default)s)')
   parser.add_argument(
-    '--dataset_root', type=str, default='data',
+    '--data_root', type=str, default='data',
     help='Default location where data is stored (default: %(default)s)')
   parser.add_argument(
     '--weights_root', type=str, default='weights',
@@ -521,15 +521,15 @@ def __len__(self):
 
 
 # Convenience function to centralize all data loaders
-def get_data_loaders(dataset, dataset_root=None, augment=False, batch_size=64, 
+def get_data_loaders(dataset, data_root=None, augment=False, batch_size=64, 
                      num_workers=8, shuffle=True, load_in_mem=False, hdf5=False,
                      pin_memory=True, drop_last=True, start_itr=0,
                      num_epochs=500, use_multiepoch_sampler=False,
                      **kwargs):
 
   # Append /FILENAME.hdf5 to root if using hdf5
-  dataset_root += '/%s' % root_dict[dataset]
-  print('Using dataset root location %s' % dataset_root)
+  data_root += '/%s' % root_dict[dataset]
+  print('Using dataset root location %s' % data_root)
 
   which_dataset = dset_dict[dataset]
   norm_mean = [0.5,0.5,0.5]
@@ -562,7 +562,7 @@ def get_data_loaders(dataset, dataset_root=None, augment=False, batch_size=64,
     train_transform = transforms.Compose(train_transform + [
                      transforms.ToTensor(),
                      transforms.Normalize(norm_mean, norm_std)])
-  train_set = which_dataset(root=dataset_root, transform=train_transform,
+  train_set = which_dataset(root=data_root, transform=train_transform,
                             load_in_mem=load_in_mem, **dataset_kwargs)
 
   # Prepare loader; the loaders list is for forward compatibility with