Mixup and prefetcher improvements

* Do mixup in custom collate fn if prefetcher enabled, reduces performance impact
* Move mixup code to own file
* Add arg to disable prefetcher
* Fix no cuda transfer when prefetcher off
* Random erasing when prefetcher off wasn't changed to match new args, fixed
* Default random erasing to off (prob = 0.) for train

data/__init__.py

@@ -3,3 +3,4 @@ from data.config import resolve_data_config
 from data.dataset import Dataset
 from data.transforms import *
 from data.loader import create_loader
+from data.mixup import mixup_target, FastCollateMixup

data/loader.py

@@ -1,6 +1,7 @@
 import torch.utils.data
 from data.transforms import *
 from data.distributed_sampler import OrderedDistributedSampler
+from data.mixup import FastCollateMixup
 
 
 def fast_collate(batch):
@@ -60,6 +61,18 @@ class PrefetchLoader:
     def sampler(self):
         return self.loader.sampler
 
+    @property
+    def mixup_enabled(self):
+        if isinstance(self.loader.collate_fn, FastCollateMixup):
+            return self.loader.collate_fn.mixup_enabled
+        else:
+            return False
+
+    @mixup_enabled.setter
+    def mixup_enabled(self, x):
+        if isinstance(self.loader.collate_fn, FastCollateMixup):
+            self.loader.collate_fn.mixup_enabled = x
+
 
 def create_loader(
         dataset,
@@ -75,6 +88,7 @@ def create_loader(
         num_workers=1,
         distributed=False,
         crop_pct=None,
+        collate_fn=None,
 ):
     if isinstance(input_size, tuple):
         img_size = input_size[-2:]
@@ -108,13 +122,16 @@ def create_loader(
             # of samples per-process, will slightly alter validation results
             sampler = OrderedDistributedSampler(dataset)
 
+    if collate_fn is None:
+        collate_fn = fast_collate if use_prefetcher else torch.utils.data.dataloader.default_collate
+
     loader = torch.utils.data.DataLoader(
         dataset,
         batch_size=batch_size,
         shuffle=sampler is None and is_training,
         num_workers=num_workers,
         sampler=sampler,
-        collate_fn=fast_collate if use_prefetcher else torch.utils.data.dataloader.default_collate,
+        collate_fn=collate_fn,
         drop_last=is_training,
     )
     if use_prefetcher:

data/mixup.py

@@ -0,0 +1,42 @@
+import numpy as np
+import torch
+
+
+def one_hot(x, num_classes, on_value=1., off_value=0., device='cuda'):
+    x = x.long().view(-1, 1)
+    return torch.full((x.size()[0], num_classes), off_value, device=device).scatter_(1, x, on_value)
+
+
+def mixup_target(target, num_classes, lam=1., smoothing=0.0, device='cuda'):
+    off_value = smoothing / num_classes
+    on_value = 1. - smoothing + off_value
+    y1 = one_hot(target, num_classes, on_value=on_value, off_value=off_value, device=device)
+    y2 = one_hot(target.flip(0), num_classes, on_value=on_value, off_value=off_value, device=device)
+    return lam*y1 + (1. - lam)*y2
+
+
+class FastCollateMixup:
+
+    def __init__(self, mixup_alpha=1., label_smoothing=0.1, num_classes=1000):
+        self.mixup_alpha = mixup_alpha
+        self.label_smoothing = label_smoothing
+        self.num_classes = num_classes
+        self.mixup_enabled = True
+
+    def __call__(self, batch):
+        batch_size = len(batch)
+        lam = 1.
+        if self.mixup_enabled:
+            lam = np.random.beta(self.mixup_alpha, self.mixup_alpha)
+
+        target = torch.tensor([b[1] for b in batch], dtype=torch.int64)
+        target = mixup_target(target, self.num_classes, lam, self.label_smoothing, device='cpu')
+
+        tensor = torch.zeros((batch_size, *batch[0][0].shape), dtype=torch.uint8)
+        for i in range(batch_size):
+            mixed = batch[i][0].astype(np.float32) * lam + \
+                    batch[batch_size - i - 1][0].astype(np.float32) * (1 - lam)
+            np.round(mixed, out=mixed)
+            tensor[i] += torch.from_numpy(mixed.astype(np.uint8))
+
+        return tensor, target

data/transforms.py

@@ -159,7 +159,7 @@ def transforms_imagenet_train(
         color_jitter=(0.4, 0.4, 0.4),
         interpolation='random',
         random_erasing=0.4,
-        random_erasing_pp=True,
+        random_erasing_mode='const',
         use_prefetcher=False,
         mean=IMAGENET_DEFAULT_MEAN,
         std=IMAGENET_DEFAULT_STD
@@ -183,7 +183,7 @@ def transforms_imagenet_train(
                 std=torch.tensor(std))
         ]
         if random_erasing > 0.:
-            tfl.append(RandomErasing(random_erasing, per_pixel=random_erasing_pp, device='cpu'))
+            tfl.append(RandomErasing(random_erasing, mode=random_erasing_mode, device='cpu'))
     return transforms.Compose(tfl)
 
 

train.py

@@ -10,7 +10,7 @@ try:
 except ImportError:
     has_apex = False
 
-from data import Dataset, create_loader, resolve_data_config
+from data import Dataset, create_loader, resolve_data_config, FastCollateMixup, mixup_target
 from models import create_model, resume_checkpoint
 from utils import *
 from loss import LabelSmoothingCrossEntropy, SparseLabelCrossEntropy
@@ -66,9 +66,9 @@ parser.add_argument('--decay-rate', '--dr', type=float, default=0.1, metavar='RA
 parser.add_argument('--sched', default='step', type=str, metavar='SCHEDULER',
                     help='LR scheduler (default: "step"')
 parser.add_argument('--drop', type=float, default=0.0, metavar='DROP',
-                    help='Dropout rate (default: 0.1)')
-parser.add_argument('--reprob', type=float, default=0.4, metavar='PCT',
-                    help='Random erase prob (default: 0.4)')
+                    help='Dropout rate (default: 0.)')
+parser.add_argument('--reprob', type=float, default=0., metavar='PCT',
+                    help='Random erase prob (default: 0.)')
 parser.add_argument('--remode', type=str, default='const',
                     help='Random erase mode (default: "const")')
 parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
@@ -109,6 +109,8 @@ parser.add_argument('--save-images', action='store_true', default=False,
                     help='save images of input bathes every log interval for debugging')
 parser.add_argument('--amp', action='store_true', default=False,
                     help='use NVIDIA amp for mixed precision training')
+parser.add_argument('--no-prefetcher', action='store_true', default=False,
+                    help='disable fast prefetcher')
 parser.add_argument('--output', default='', type=str, metavar='PATH',
                     help='path to output folder (default: none, current dir)')
 parser.add_argument('--eval-metric', default='prec1', type=str, metavar='EVAL_METRIC',
@@ -119,6 +121,7 @@ parser.add_argument("--local_rank", default=0, type=int)
 def main():
     args = parser.parse_args()
 
+    args.prefetcher = not args.no_prefetcher
     args.distributed = False
     if 'WORLD_SIZE' in os.environ:
         args.distributed = int(os.environ['WORLD_SIZE']) > 1
@@ -130,6 +133,7 @@ def main():
     args.world_size = 1
     r = -1
     if args.distributed:
+        args.num_gpu = 1
         args.device = 'cuda:%d' % args.local_rank
         torch.cuda.set_device(args.local_rank)
         torch.distributed.init_process_group(backend='nccl',
@@ -216,12 +220,16 @@ def main():
         exit(1)
     dataset_train = Dataset(train_dir)
 
+    collate_fn = None
+    if args.prefetcher and args.mixup > 0:
+        collate_fn = FastCollateMixup(args.mixup, args.smoothing, args.num_classes)
+
     loader_train = create_loader(
         dataset_train,
         input_size=data_config['input_size'],
         batch_size=args.batch_size,
         is_training=True,
-        use_prefetcher=True,
+        use_prefetcher=args.prefetcher,
         rand_erase_prob=args.reprob,
         rand_erase_mode=args.remode,
         interpolation='random',  # FIXME cleanly resolve this? data_config['interpolation'],
@@ -229,6 +237,7 @@ def main():
         std=data_config['std'],
         num_workers=args.workers,
         distributed=args.distributed,
+        collate_fn=collate_fn,
     )
 
     eval_dir = os.path.join(args.data, 'validation')
@@ -242,7 +251,7 @@ def main():
         input_size=data_config['input_size'],
         batch_size=4 * args.batch_size,
         is_training=False,
-        use_prefetcher=True,
+        use_prefetcher=args.prefetcher,
         interpolation=data_config['interpolation'],
         mean=data_config['mean'],
         std=data_config['std'],
@@ -309,6 +318,10 @@ def train_epoch(
         epoch, model, loader, optimizer, loss_fn, args,
         lr_scheduler=None, saver=None, output_dir='', use_amp=False):
 
+    if args.prefetcher and args.mixup > 0 and loader.mixup_enabled:
+        if args.mixup_off_epoch and epoch >= args.mixup_off_epoch:
+            loader.mixup_enabled = False
+
     batch_time_m = AverageMeter()
     data_time_m = AverageMeter()
     losses_m = AverageMeter()
@@ -321,13 +334,15 @@ def train_epoch(
     for batch_idx, (input, target) in enumerate(loader):
         last_batch = batch_idx == last_idx
         data_time_m.update(time.time() - end)
-
-        if args.mixup > 0.:
-            lam = 1.
-            if not args.mixup_off_epoch or epoch < args.mixup_off_epoch:
-                lam = np.random.beta(args.mixup, args.mixup)
-            input.mul_(lam).add_(1 - lam, input.flip(0))
-            target = mixup_target(target, args.num_classes, lam, args.smoothing)
+        if not args.prefetcher:
+            input = input.cuda()
+            target = target.cuda()
+            if args.mixup > 0.:
+                lam = 1.
+                if not args.mixup_off_epoch or epoch < args.mixup_off_epoch:
+                    lam = np.random.beta(args.mixup, args.mixup)
+                input.mul_(lam).add_(1 - lam, input.flip(0))
+                target = mixup_target(target, args.num_classes, lam, args.smoothing)
 
         output = model(input)
 

utils.py

@@ -140,19 +140,6 @@ def accuracy(output, target, topk=(1,)):
     return res
 
 
-def one_hot(x, num_classes, on_value=1., off_value=0., device='cuda'):
-    x = x.long().view(-1, 1)
-    return torch.full((x.size()[0], num_classes), off_value, device=device).scatter_(1, x, on_value)
-
-
-def mixup_target(target, num_classes, lam=1., smoothing=0.0):
-    off_value = smoothing / num_classes
-    on_value = 1. - smoothing + off_value
-    y1 = one_hot(target, num_classes, on_value=on_value, off_value=off_value)
-    y2 = one_hot(target.flip(0), num_classes, on_value=on_value, off_value=off_value)
-    return lam*y1 + (1. - lam)*y2
-
-
 def get_outdir(path, *paths, inc=False):
     outdir = os.path.join(path, *paths)
     if not os.path.exists(outdir):