Final cleanup of mixup/cutmix. Element/batch modes working with both collate (prefetcher active) and without prefetcher.

timm/data/__init__.py

@@ -4,7 +4,7 @@ from .dataset import Dataset, DatasetTar, AugMixDataset
 from .transforms import *
 from .loader import create_loader
 from .transforms_factory import create_transform
-from .mixup import mix_batch, FastCollateMixup
+from .mixup import Mixup, FastCollateMixup
 from .auto_augment import RandAugment, AutoAugment, rand_augment_ops, auto_augment_policy,\
     rand_augment_transform, auto_augment_transform
 from .real_labels import RealLabelsImagenet

timm/data/mixup.py

@@ -10,11 +10,8 @@ CutMix: https://github.com/clovaai/CutMix-PyTorch
 
 Hacked together by / Copyright 2020 Ross Wightman
 """
-
 import numpy as np
 import torch
-import math
-import numbers
 
 
 def one_hot(x, num_classes, on_value=1., off_value=0., device='cuda'):
@@ -30,20 +27,21 @@ def mixup_target(target, num_classes, lam=1., smoothing=0.0, device='cuda'):
     return y1 * lam + y2 * (1. - lam)
 
 
-def mixup_batch(input, target, alpha=0.2, num_classes=1000, smoothing=0.1, disable=False):
+def rand_bbox(img_shape, lam, margin=0., count=None):
-    lam = 1.
+    """ Standard CutMix bounding-box
-    if not disable:
+    Generates a random square bbox based on lambda value. This impl includes
-        lam = np.random.beta(alpha, alpha)
+    support for enforcing a border margin as percent of bbox dimensions.
-    input = input.mul(lam).add_(1 - lam, input.flip(0))
-    target = mixup_target(target, num_classes, lam, smoothing)
-    return input, target
-
 
-def rand_bbox(size, lam, border=0., count=None):
+    Args:
-    ratio = math.sqrt(1 - lam)
+        img_shape (tuple): Image shape as tuple
-    img_h, img_w = size[-2:]
+        lam (float): Cutmix lambda value
+        margin (float): Percentage of bbox dimension to enforce as margin (reduce amount of box outside image)
+        count (int): Number of bbox to generate
+    """
+    ratio = np.sqrt(1 - lam)
+    img_h, img_w = img_shape[-2:]
     cut_h, cut_w = int(img_h * ratio), int(img_w * ratio)
-    margin_y, margin_x = int(border * cut_h), int(border * cut_w)
+    margin_y, margin_x = int(margin * cut_h), int(margin * cut_w)
     cy = np.random.randint(0 + margin_y, img_h - margin_y, size=count)
     cx = np.random.randint(0 + margin_x, img_w - margin_x, size=count)
     yl = np.clip(cy - cut_h // 2, 0, img_h)
@@ -53,9 +51,20 @@ def rand_bbox(size, lam, border=0., count=None):
     return yl, yh, xl, xh
 
 
-def rand_bbox_minmax(size, minmax, count=None):
+def rand_bbox_minmax(img_shape, minmax, count=None):
+    """ Min-Max CutMix bounding-box
+    Inspired by Darknet cutmix impl, generates a random rectangular bbox
+    based on min/max percent values applied to each dimension of the input image.
+
+    Typical defaults for minmax are usually in the  .2-.3 for min and .8-.9 range for max.
+
+    Args:
+        img_shape (tuple): Image shape as tuple
+        minmax (tuple or list): Min and max bbox ratios (as percent of image size)
+        count (int): Number of bbox to generate
+    """
     assert len(minmax) == 2
-    img_h, img_w = size[-2:]
+    img_h, img_w = img_shape[-2:]
     cut_h = np.random.randint(int(img_h * minmax[0]), int(img_h * minmax[1]), size=count)
     cut_w = np.random.randint(int(img_w * minmax[0]), int(img_w * minmax[1]), size=count)
     yl = np.random.randint(0, img_h - cut_h, size=count)
@@ -66,6 +75,8 @@ def rand_bbox_minmax(size, minmax, count=None):
 
 
 def cutmix_bbox_and_lam(img_shape, lam, ratio_minmax=None, correct_lam=True, count=None):
+    """ Generate bbox and apply lambda correction.
+    """
     if ratio_minmax is not None:
         yl, yu, xl, xu = rand_bbox_minmax(img_shape, ratio_minmax, count=count)
     else:
@@ -76,52 +87,22 @@ def cutmix_bbox_and_lam(img_shape, lam, ratio_minmax=None, correct_lam=True, cou
     return (yl, yu, xl, xu), lam
 
 
-def cutmix_batch(input, target, alpha=0.2, num_classes=1000, smoothing=0.1, disable=False, correct_lam=False):
+class Mixup:
-    lam = 1.
+    """ Mixup/Cutmix that applies different params to each element or whole batch
-    if not disable:
-        lam = np.random.beta(alpha, alpha)
-    if lam != 1:
-        yl, yh, xl, xh = rand_bbox(input.size(), lam)
-        input[:, :, yl:yh, xl:xh] = input.flip(0)[:, :, yl:yh, xl:xh]
-        if correct_lam:
-            lam = 1. - (yh - yl) * (xh - xl) / float(input.shape[-2] * input.shape[-1])
-    target = mixup_target(target, num_classes, lam, smoothing)
-    return input, target
-
-
-def mix_batch(
-        input, target, mixup_alpha=0.2, cutmix_alpha=0., prob=1.0, switch_prob=.5,
-        num_classes=1000, smoothing=0.1, disable=False):
-    # FIXME test this version
-    if np.random.rand() > prob:
-        return input, target
-    use_cutmix = cutmix_alpha > 0. and np.random.rand() <= switch_prob
-    if use_cutmix:
-        return cutmix_batch(input, target, cutmix_alpha, num_classes, smoothing, disable)
-    else:
-        return mixup_batch(input, target, mixup_alpha, num_classes, smoothing, disable)
-
-
-class FastCollateMixup:
-    """Fast Collate Mixup/Cutmix that applies different params to each element or whole batch
-
-    NOTE once experiments are done, one of the three variants will remain with this class name
 
+    Args:
+        mixup_alpha (float): mixup alpha value, mixup is active if > 0.
+        cutmix_alpha (float): cutmix alpha value, cutmix is active if > 0.
+        cutmix_minmax (List[float]): cutmix min/max image ratio, cutmix is active and uses this vs alpha if not None.
+        prob (float): probability of applying mixup or cutmix per batch or element
+        switch_prob (float): probability of switching to cutmix instead of mixup when both are active
+        elementwise (bool): apply mixup/cutmix params per batch element instead of per batch
+        correct_lam (bool): apply lambda correction when cutmix bbox clipped by image borders
+        label_smoothing (float): apply label smoothing to the mixed target tensor
+        num_classes (int): number of classes for target
     """
     def __init__(self, mixup_alpha=1., cutmix_alpha=0., cutmix_minmax=None, prob=1.0, switch_prob=0.5,
                  elementwise=False, correct_lam=True, label_smoothing=0.1, num_classes=1000):
-        """
-
-        Args:
-            mixup_alpha (float): mixup alpha value, mixup is active if > 0.
-            cutmix_alpha (float): cutmix alpha value, cutmix is active if > 0.
-            cutmix_minmax (float): cutmix min/max image ratio, cutmix is active and uses this vs alpha if not None
-            prob (float): probability of applying mixup or cutmix per batch or element
-            switch_prob (float): probability of using cutmix instead of mixup when both active
-            elementwise (bool): apply mixup/cutmix params per batch element instead of per batch
-            label_smoothing (float):
-            num_classes (int):
-        """
         self.mixup_alpha = mixup_alpha
         self.cutmix_alpha = cutmix_alpha
         self.cutmix_minmax = cutmix_minmax
@@ -129,7 +110,7 @@ class FastCollateMixup:
             assert len(self.cutmix_minmax) == 2
             # force cutmix alpha == 1.0 when minmax active to keep logic simple & safe
             self.cutmix_alpha = 1.0
-        self.prob = prob
+        self.mix_prob = prob
         self.switch_prob = switch_prob
         self.label_smoothing = label_smoothing
         self.num_classes = num_classes
@@ -137,10 +118,9 @@ class FastCollateMixup:
         self.correct_lam = correct_lam  # correct lambda based on clipped area for cutmix
         self.mixup_enabled = True  # set to false to disable mixing (intended tp be set by train loop)
 
-    def _mix_elem(self, output, batch):
+    def _params_per_elem(self, batch_size):
-        batch_size = len(batch)
+        lam = np.ones(batch_size, dtype=np.float32)
-        lam_out = np.ones(batch_size, dtype=np.float32)
+        use_cutmix = np.zeros(batch_size, dtype=np.bool)
-        use_cutmix = np.zeros(batch_size).astype(np.bool)
         if self.mixup_enabled:
             if self.mixup_alpha > 0. and self.cutmix_alpha > 0.:
                 use_cutmix = np.random.rand(batch_size) < self.switch_prob
@@ -151,35 +131,17 @@ class FastCollateMixup:
             elif self.mixup_alpha > 0.:
                 lam_mix = np.random.beta(self.mixup_alpha, self.mixup_alpha, size=batch_size)
             elif self.cutmix_alpha > 0.:
-                use_cutmix = np.ones(batch_size).astype(np.bool)
+                use_cutmix = np.ones(batch_size, dtype=np.bool)
                 lam_mix = np.random.beta(self.cutmix_alpha, self.cutmix_alpha, size=batch_size)
             else:
                 assert False, "One of mixup_alpha > 0., cutmix_alpha > 0., cutmix_minmax not None should be true."
-            lam_out = np.where(np.random.rand(batch_size) < self.prob, lam_mix.astype(np.float32), lam_out)
+            lam = np.where(np.random.rand(batch_size) < self.mix_prob, lam_mix.astype(np.float32), lam)
-
+        return lam, use_cutmix
-        for i in range(batch_size):
-            j = batch_size - i - 1
-            lam = lam_out[i]
-            mixed = batch[i][0]
-            if lam != 1.:
-                if use_cutmix[i]:
-                    mixed = mixed.copy()
-                    (yl, yh, xl, xh), lam = cutmix_bbox_and_lam(
-                        output.shape, lam, ratio_minmax=self.cutmix_minmax, correct_lam=self.correct_lam)
-                    mixed[:, yl:yh, xl:xh] = batch[j][0][:, yl:yh, xl:xh]
-                    lam_out[i] = lam
-                else:
-                    mixed = mixed.astype(np.float32) * lam + batch[j][0].astype(np.float32) * (1 - lam)
-                    lam_out[i] = lam
-                    np.round(mixed, out=mixed)
-            output[i] += torch.from_numpy(mixed.astype(np.uint8))
-        return torch.tensor(lam_out).unsqueeze(1)
 
-    def _mix_batch(self, output, batch):
+    def _params_per_batch(self):
-        batch_size = len(batch)
         lam = 1.
         use_cutmix = False
-        if self.mixup_enabled and np.random.rand() < self.prob:
+        if self.mixup_enabled and np.random.rand() < self.mix_prob:
             if self.mixup_alpha > 0. and self.cutmix_alpha > 0.:
                 use_cutmix = np.random.rand() < self.switch_prob
                 lam_mix = np.random.beta(self.cutmix_alpha, self.cutmix_alpha) if use_cutmix else \
@@ -192,17 +154,84 @@ class FastCollateMixup:
             else:
                 assert False, "One of mixup_alpha > 0., cutmix_alpha > 0., cutmix_minmax not None should be true."
             lam = float(lam_mix)
+        return lam, use_cutmix
 
+    def _mix_elem(self, x):
+        batch_size = len(x)
+        lam_batch, use_cutmix = self._params_per_elem(batch_size)
+        x_orig = x.clone()  # need to keep an unmodified original for mixing source
+        for i in range(batch_size):
+            j = batch_size - i - 1
+            lam = lam_batch[i]
+            if lam != 1.:
+                if use_cutmix[i]:
+                    (yl, yh, xl, xh), lam = cutmix_bbox_and_lam(
+                        x[i].shape, lam, ratio_minmax=self.cutmix_minmax, correct_lam=self.correct_lam)
+                    x[i][:, yl:yh, xl:xh] = x_orig[j][:, yl:yh, xl:xh]
+                    lam_batch[i] = lam
+                else:
+                    x[i] = x[i] * lam + x_orig[j] * (1 - lam)
+        return torch.tensor(lam_batch, device=x.device, dtype=x.dtype).unsqueeze(1)
+
+    def _mix_batch(self, x):
+        lam, use_cutmix = self._params_per_batch()
+        if lam == 1.:
+            return 1.
         if use_cutmix:
             (yl, yh, xl, xh), lam = cutmix_bbox_and_lam(
-                output.shape, lam, ratio_minmax=self.cutmix_minmax, correct_lam=self.correct_lam)
+                x.shape, lam, ratio_minmax=self.cutmix_minmax, correct_lam=self.correct_lam)
+            x[:, :, yl:yh, xl:xh] = x.flip(0)[:, :, yl:yh, xl:xh]
+        else:
+            x_flipped = x.flip(0).mul_(1. - lam)
+            x.mul_(lam).add_(x_flipped)
+        return lam
+
+    def __call__(self, x, target):
+        assert len(x) % 2 == 0, 'Batch size should be even when using this'
+        lam = self._mix_elem(x) if self.elementwise else self._mix_batch(x)
+        target = mixup_target(target, self.num_classes, lam, self.label_smoothing)
+        return x, target
+
+
+class FastCollateMixup(Mixup):
+    """ Fast Collate w/ Mixup/Cutmix that applies different params to each element or whole batch
 
+    A Mixup impl that's performed while collating the batches.
+    """
+
+    def _mix_elem_collate(self, output, batch):
+        batch_size = len(batch)
+        lam_batch, use_cutmix = self._params_per_elem(batch_size)
         for i in range(batch_size):
             j = batch_size - i - 1
+            lam = lam_batch[i]
             mixed = batch[i][0]
             if lam != 1.:
-                if use_cutmix:
+                if use_cutmix[i]:
                     mixed = mixed.copy()
+                    (yl, yh, xl, xh), lam = cutmix_bbox_and_lam(
+                        output.shape, lam, ratio_minmax=self.cutmix_minmax, correct_lam=self.correct_lam)
+                    mixed[:, yl:yh, xl:xh] = batch[j][0][:, yl:yh, xl:xh]
+                    lam_batch[i] = lam
+                else:
+                    mixed = mixed.astype(np.float32) * lam + batch[j][0].astype(np.float32) * (1 - lam)
+                    lam_batch[i] = lam
+                    np.round(mixed, out=mixed)
+            output[i] += torch.from_numpy(mixed.astype(np.uint8))
+        return torch.tensor(lam_batch).unsqueeze(1)
+
+    def _mix_batch_collate(self, output, batch):
+        batch_size = len(batch)
+        lam, use_cutmix = self._params_per_batch()
+        if use_cutmix:
+            (yl, yh, xl, xh), lam = cutmix_bbox_and_lam(
+                output.shape, lam, ratio_minmax=self.cutmix_minmax, correct_lam=self.correct_lam)
+        for i in range(batch_size):
+            j = batch_size - i - 1
+            mixed = batch[i][0]
+            if lam != 1.:
+                if use_cutmix:
+                    mixed = mixed.copy()  # don't want to modify the original while iterating
                     mixed[:, yl:yh, xl:xh] = batch[j][0][:, yl:yh, xl:xh]
                 else:
                     mixed = mixed.astype(np.float32) * lam + batch[j][0].astype(np.float32) * (1 - lam)
@@ -210,16 +239,15 @@ class FastCollateMixup:
             output[i] += torch.from_numpy(mixed.astype(np.uint8))
         return lam
 
-    def __call__(self, batch):
+    def __call__(self, batch, _=None):
         batch_size = len(batch)
         assert batch_size % 2 == 0, 'Batch size should be even when using this'
         output = torch.zeros((batch_size, *batch[0][0].shape), dtype=torch.uint8)
         if self.elementwise:
-            lam = self._mix_elem(output, batch)
+            lam = self._mix_elem_collate(output, batch)
         else:
-            lam = self._mix_batch(output, batch)
+            lam = self._mix_batch_collate(output, batch)
         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')
-
         return output, target
 

train.py

@@ -28,7 +28,7 @@ except ImportError:
     from torch.nn.parallel import DistributedDataParallel as DDP
     has_apex = False
 
-from timm.data import Dataset, create_loader, resolve_data_config, FastCollateMixup, mix_batch, AugMixDataset
+from timm.data import Dataset, create_loader, resolve_data_config, Mixup, FastCollateMixup, AugMixDataset
 from timm.models import create_model, resume_checkpoint, convert_splitbn_model
 from timm.utils import *
 from timm.loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy, JsdCrossEntropy
@@ -398,12 +398,18 @@ def main():
     dataset_train = Dataset(train_dir)
 
     collate_fn = None
-    if args.prefetcher and (args.mixup > 0 or args.cutmix > 0. or args.cutmix_minmax is not None):
+    mixup_fn = None
-        assert not num_aug_splits  # collate conflict (need to support deinterleaving in collate mixup)
+    mixup_active = args.mixup > 0 or args.cutmix > 0. or args.cutmix_minmax is not None
-        collate_fn = FastCollateMixup(
+    if mixup_active:
+        mixup_args = dict(
             mixup_alpha=args.mixup, cutmix_alpha=args.cutmix, cutmix_minmax=args.cutmix_minmax,
             prob=args.mixup_prob, switch_prob=args.mixup_switch_prob, elementwise=args.mixup_elem,
             label_smoothing=args.smoothing, num_classes=args.num_classes)
+        if args.prefetcher:
+            assert not num_aug_splits  # collate conflict (need to support deinterleaving in collate mixup)
+            collate_fn = FastCollateMixup(**mixup_args)
+        else:
+            mixup_fn = Mixup(**mixup_args)
 
     if num_aug_splits > 1:
         dataset_train = AugMixDataset(dataset_train, num_splits=num_aug_splits)
@@ -465,17 +471,14 @@ def main():
     if args.jsd:
         assert num_aug_splits > 1  # JSD only valid with aug splits set
         train_loss_fn = JsdCrossEntropy(num_splits=num_aug_splits, smoothing=args.smoothing).cuda()
-        validate_loss_fn = nn.CrossEntropyLoss().cuda()
+    elif mixup_active:
-    elif args.mixup > 0.:
+        # smoothing is handled with mixup target transform
-        # smoothing is handled with mixup label transform
         train_loss_fn = SoftTargetCrossEntropy().cuda()
-        validate_loss_fn = nn.CrossEntropyLoss().cuda()
     elif args.smoothing:
         train_loss_fn = LabelSmoothingCrossEntropy(smoothing=args.smoothing).cuda()
-        validate_loss_fn = nn.CrossEntropyLoss().cuda()
     else:
         train_loss_fn = nn.CrossEntropyLoss().cuda()
-        validate_loss_fn = train_loss_fn
+    validate_loss_fn = nn.CrossEntropyLoss().cuda()
 
     eval_metric = args.eval_metric
     best_metric = None
@@ -503,7 +506,7 @@ def main():
             train_metrics = train_epoch(
                 epoch, model, loader_train, optimizer, train_loss_fn, args,
                 lr_scheduler=lr_scheduler, saver=saver, output_dir=output_dir,
-                use_amp=use_amp, model_ema=model_ema)
+                use_amp=use_amp, model_ema=model_ema, mixup_fn=mixup_fn)
 
             if args.distributed and args.dist_bn in ('broadcast', 'reduce'):
                 if args.local_rank == 0:
@@ -543,11 +546,13 @@ def main():
 
 def train_epoch(
         epoch, model, loader, optimizer, loss_fn, args,
-        lr_scheduler=None, saver=None, output_dir='', use_amp=False, model_ema=None):
+        lr_scheduler=None, saver=None, output_dir='', use_amp=False, model_ema=None, mixup_fn=None):
 
-    if args.prefetcher and args.mixup > 0 and loader.mixup_enabled:
+    if args.mixup_off_epoch and epoch >= args.mixup_off_epoch:
-        if args.mixup_off_epoch and epoch >= args.mixup_off_epoch:
+        if args.prefetcher and loader.mixup_enabled:
             loader.mixup_enabled = False
+        elif mixup_fn is not None:
+            mixup_fn.mixup_enabled = False
 
     batch_time_m = AverageMeter()
     data_time_m = AverageMeter()
@@ -563,12 +568,8 @@ def train_epoch(
         data_time_m.update(time.time() - end)
         if not args.prefetcher:
             input, target = input.cuda(), target.cuda()
-            if args.mixup > 0.:
+            if mixup_fn is not None:
-                input, target = mix_batch(
+                input, target = mixup_fn(input, target)
-                    input, target,
-                    mixup_alpha=args.mixup, cutmix_alpha=args.cutmix, prob=args.mixup_prob,
-                    switch_prob=args.mixup_switch_prob, num_classes=args.num_classes, smoothing=args.smoothing,
-                    disable=args.mixup_off_epoch and epoch >= args.mixup_off_epoch)
 
         output = model(input)