add gpu resnext50

model_zoo/official/cv/resnext50/README.md

@@ -90,10 +90,15 @@ sh run_standalone_train.sh DEVICE_ID DATA_PATH
 #### Launch
 
 ```bash
-# distributed training example(8p)
+# distributed training example(8p) for Ascend
 sh scripts/run_distribute_train.sh MINDSPORE_HCCL_CONFIG_PATH /dataset/train
-# standalone training example
+# standalone training example for Ascend
 sh scripts/run_standalone_train.sh 0 /dataset/train
+
+# distributed training example(8p) for GPU
+sh scripts/run_distribute_train_for_gpu.sh /dataset/train
+# standalone training example for GPU
+sh scripts/run_standalone_train_for_gpu.sh 0 /dataset/train
 ```
 
 #### Result
@@ -106,14 +111,15 @@ You can find checkpoint file together with result in log.
 
 ```
 # Evaluation
-sh run_eval.sh DEVICE_ID DATA_PATH PRETRAINED_CKPT_PATH
+sh run_eval.sh DEVICE_ID DATA_PATH PRETRAINED_CKPT_PATH PLATFORM
 ```
+PLATFORM is Ascend or GPU, default is Ascend.
 
 #### Launch
 
 ```bash
 # Evaluation with checkpoint
-sh scripts/run_eval.sh 0 /opt/npu/datasets/classification/val /resnext50_100.ckpt
+sh scripts/run_eval.sh 0 /opt/npu/datasets/classification/val /resnext50_100.ckpt Ascend
 ```
 
 > checkpoint can be produced in training process.

model_zoo/official/cv/resnext50/eval.py

@@ -29,15 +29,11 @@ from mindspore.ops import functional as F
 from mindspore.common import dtype as mstype
 
 from src.utils.logging import get_logger
+from src.utils.auto_mixed_precision import auto_mixed_precision
 from src.image_classification import get_network
 from src.dataset import classification_dataset
 from src.config import config
 
-devid = int(os.getenv('DEVICE_ID'))
-context.set_context(mode=context.GRAPH_MODE, enable_auto_mixed_precision=True,
-                    device_target="Ascend", save_graphs=False, device_id=devid)
-
-
 
 class ParameterReduce(nn.Cell):
     """ParameterReduce"""
@@ -56,6 +52,7 @@ class ParameterReduce(nn.Cell):
 def parse_args(cloud_args=None):
     """parse_args"""
     parser = argparse.ArgumentParser('mindspore classification test')
+    parser.add_argument('--platform', type=str, default='Ascend', choices=('Ascend', 'GPU'), help='run platform')
 
     # dataset related
     parser.add_argument('--data_dir', type=str, default='/opt/npu/datasets/classification/val', help='eval data dir')
@@ -108,12 +105,25 @@ def merge_args(args, cloud_args):
 def test(cloud_args=None):
     """test"""
     args = parse_args(cloud_args)
+    context.set_context(mode=context.GRAPH_MODE, enable_auto_mixed_precision=True,
+                        device_target=args.platform, save_graphs=False)
+    if os.getenv('DEVICE_ID', "not_set").isdigit():
+        context.set_context(device_id=int(os.getenv('DEVICE_ID')))
 
     # init distributed
     if args.is_distributed:
-        init()
+        if args.platform == "Ascend":
+            init()
+        elif args.platform == "GPU":
+            init("nccl")
         args.rank = get_rank()
         args.group_size = get_group_size()
+        parallel_mode = ParallelMode.DATA_PARALLEL
+        context.set_auto_parallel_context(parallel_mode=parallel_mode, device_num=args.group_size,
+                                          parameter_broadcast=True, mirror_mean=True)
+    else:
+        args.rank = 0
+        args.group_size = 1
 
     args.outputs_dir = os.path.join(args.log_path,
                                     datetime.datetime.now().strftime('%Y-%m-%d_time_%H_%M_%S'))
@@ -140,7 +150,7 @@ def test(cloud_args=None):
                                             max_epoch=1, rank=args.rank, group_size=args.group_size,
                                             mode='eval')
         eval_dataloader = de_dataset.create_tuple_iterator()
-        network = get_network(args.backbone, args.num_classes)
+        network = get_network(args.backbone, args.num_classes, platform=args.platform)
         if network is None:
             raise NotImplementedError('not implement {}'.format(args.backbone))
 
@@ -157,12 +167,13 @@ def test(cloud_args=None):
         load_param_into_net(network, param_dict_new)
         args.logger.info('load model {} success'.format(model))
 
-        # must add
-        network.add_flags_recursive(fp16=True)
-
         img_tot = 0
         top1_correct = 0
         top5_correct = 0
+        if args.platform == "Ascend":
+            network.to_float(mstype.float16)
+        else:
+            auto_mixed_precision(network)
         network.set_train(False)
         t_end = time.time()
         it = 0

model_zoo/official/cv/resnext50/scripts/run_distribute_train_for_gpu.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+DATA_DIR=$1
+export RANK_SIZE=8
+PATH_CHECKPOINT=""
+if [ $# == 2 ]
+then
+	PATH_CHECKPOINT=$2
+fi
+
+mpirun --allow-run-as-root -n $RANK_SIZE \
+	python train.py  \
+    --is_distribute=1 \
+    --platform="GPU" \
+    --pretrained=$PATH_CHECKPOINT \
+    --data_dir=$DATA_DIR > log.txt 2>&1 &

model_zoo/official/cv/resnext50/scripts/run_eval.sh

@@ -14,11 +14,16 @@
 # limitations under the License.
 # ============================================================================
 
-DEVICE_ID=$1
+export DEVICE_ID=$1
 DATA_DIR=$2
 PATH_CHECKPOINT=$3
+PLATFORM=Ascend
+if [ $# == 4 ]
+then
+  PLATFORM=$4
+fi
 
 python eval.py  \
-    --device_id=$DEVICE_ID \
     --pretrained=$PATH_CHECKPOINT \
+    --platform=$PLATFORM \
     --data_dir=$DATA_DIR > log.txt 2>&1 &

model_zoo/official/cv/resnext50/scripts/run_standalone_train.sh

@@ -14,7 +14,7 @@
 # limitations under the License.
 # ============================================================================
 
-DEVICE_ID=$1
+export DEVICE_ID=$1
 DATA_DIR=$2
 PATH_CHECKPOINT=""
 if [ $# == 3 ]

model_zoo/official/cv/resnext50/scripts/run_standalone_train_for_gpu.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+export DEVICE_ID=$1
+DATA_DIR=$2
+PATH_CHECKPOINT=""
+if [ $# == 3 ]
+then
+  PATH_CHECKPOINT=$3
+fi
+
+python train.py  \
+    --is_distribute=0 \
+    --pretrained=$PATH_CHECKPOINT \
+    --platform="GPU" \
+    --data_dir=$DATA_DIR > log.txt 2>&1 &
+

model_zoo/official/cv/resnext50/src/backbone/resnet.py

@@ -87,7 +87,8 @@ class BasicBlock(nn.Cell):
     """
     expansion = 1
 
-    def __init__(self, in_channels, out_channels, stride=1, down_sample=None, use_se=False, **kwargs):
+    def __init__(self, in_channels, out_channels, stride=1, down_sample=None, use_se=False,
+                 platform="Ascend", **kwargs):
         super(BasicBlock, self).__init__()
         self.conv1 = conv3x3(in_channels, out_channels, stride=stride)
         self.bn1 = nn.BatchNorm2d(out_channels)
@@ -142,7 +143,7 @@ class Bottleneck(nn.Cell):
     expansion = 4
 
     def __init__(self, in_channels, out_channels, stride=1, down_sample=None,
-                 base_width=64, groups=1, use_se=False, **kwargs):
+                 base_width=64, groups=1, use_se=False, platform="Ascend", **kwargs):
         super(Bottleneck, self).__init__()
 
         width = int(out_channels * (base_width / 64.0)) * groups
@@ -153,7 +154,11 @@ class Bottleneck(nn.Cell):
 
         self.conv3x3s = nn.CellList()
 
-        self.conv2 = GroupConv(width, width, 3, stride, pad=1, groups=groups)
+        if platform == "GPU":
+            self.conv2 = nn.Conv2d(width, width, 3, stride, pad_mode='pad', padding=1, group=groups)
+        else:
+            self.conv2 = GroupConv(width, width, 3, stride, pad=1, groups=groups)
+
         self.op_split = Split(axis=1, output_num=self.groups)
         self.op_concat = Concat(axis=1)
 
@@ -211,7 +216,7 @@ class ResNet(nn.Cell):
     Examples:
         >>>ResNet()
     """
-    def __init__(self, block, layers, width_per_group=64, groups=1, use_se=False):
+    def __init__(self, block, layers, width_per_group=64, groups=1, use_se=False, platform="Ascend"):
         super(ResNet, self).__init__()
         self.in_channels = 64
         self.groups = groups
@@ -222,10 +227,10 @@ class ResNet(nn.Cell):
         self.relu = P.ReLU()
         self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, pad_mode='same')
 
-        self.layer1 = self._make_layer(block, 64, layers[0], use_se=use_se)
-        self.layer2 = self._make_layer(block, 128, layers[1], stride=2, use_se=use_se)
-        self.layer3 = self._make_layer(block, 256, layers[2], stride=2, use_se=use_se)
-        self.layer4 = self._make_layer(block, 512, layers[3], stride=2, use_se=use_se)
+        self.layer1 = self._make_layer(block, 64, layers[0], use_se=use_se, platform=platform)
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=2, use_se=use_se, platform=platform)
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=2, use_se=use_se, platform=platform)
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=2, use_se=use_se, platform=platform)
 
         self.out_channels = 512 * block.expansion
         self.cast = P.Cast()
@@ -242,7 +247,7 @@ class ResNet(nn.Cell):
 
         return x
 
-    def _make_layer(self, block, out_channels, blocks_num, stride=1, use_se=False):
+    def _make_layer(self, block, out_channels, blocks_num, stride=1, use_se=False, platform="Ascend"):
         """_make_layer"""
         down_sample = None
         if stride != 1 or self.in_channels != out_channels * block.expansion:
@@ -257,11 +262,12 @@ class ResNet(nn.Cell):
                             down_sample=down_sample,
                             base_width=self.base_width,
                             groups=self.groups,
-                            use_se=use_se))
+                            use_se=use_se,
+                            platform=platform))
         self.in_channels = out_channels * block.expansion
         for _ in range(1, blocks_num):
-            layers.append(block(self.in_channels, out_channels,
-                                base_width=self.base_width, groups=self.groups, use_se=use_se))
+            layers.append(block(self.in_channels, out_channels, base_width=self.base_width,
+                                groups=self.groups, use_se=use_se, platform=platform))
 
         return nn.SequentialCell(layers)
 
@@ -269,5 +275,5 @@ class ResNet(nn.Cell):
         return self.out_channels
 
 
-def resnext50():
-    return ResNet(Bottleneck, [3, 4, 6, 3], width_per_group=4, groups=32)
+def resnext50(platform="Ascend"):
+    return ResNet(Bottleneck, [3, 4, 6, 3], width_per_group=4, groups=32, platform=platform)

model_zoo/official/cv/resnext50/src/config.py

@@ -36,7 +36,8 @@ config = ed({
     "label_smooth": 1,
     "label_smooth_factor": 0.1,
 
-    "ckpt_interval": 1250,
+    "ckpt_interval": 5,
+    "ckpt_save_max": 5,
     "ckpt_path": 'outputs/',
     "is_save_on_master": 1,
 

model_zoo/official/cv/resnext50/src/dataset.py

@@ -143,8 +143,10 @@ def classification_dataset(data_dir, image_size, per_batch_size, max_epoch, rank
         de_dataset = de.GeneratorDataset(dataset, ["image", "label"], sampler=sampler)
         de_dataset.set_dataset_size(len(sampler))
 
-    de_dataset = de_dataset.map(input_columns="image", num_parallel_workers=8, operations=transform_img)
-    de_dataset = de_dataset.map(input_columns="label", num_parallel_workers=8, operations=transform_label)
+    de_dataset = de_dataset.map(input_columns="image", num_parallel_workers=num_parallel_workers,
+                                operations=transform_img)
+    de_dataset = de_dataset.map(input_columns="label", num_parallel_workers=num_parallel_workers,
+                                operations=transform_label)
 
     columns_to_project = ["image", "label"]
     de_dataset = de_dataset.project(columns=columns_to_project)

model_zoo/official/cv/resnext50/src/image_classification.py

@@ -50,9 +50,9 @@ class Resnet(ImageClassificationNetwork):
     Returns:
         Resnet.
     """
-    def __init__(self, backbone_name, num_classes):
+    def __init__(self, backbone_name, num_classes, platform="Ascend"):
         self.backbone_name = backbone_name
-        backbone = backbones.__dict__[self.backbone_name]()
+        backbone = backbones.__dict__[self.backbone_name](platform=platform)
         out_channels = backbone.get_out_channels()
         head = heads.CommonHead(num_classes=num_classes, out_channels=out_channels)
         super(Resnet, self).__init__(backbone, head)
@@ -79,7 +79,7 @@ class Resnet(ImageClassificationNetwork):
 
 
 
-def get_network(backbone_name, num_classes):
+def get_network(backbone_name, num_classes, platform="Ascend"):
     if backbone_name in ['resnext50']:
-        return Resnet(backbone_name, num_classes)
+        return Resnet(backbone_name, num_classes, platform)
     return None

model_zoo/official/cv/resnext50/src/utils/auto_mixed_precision.py

@@ -0,0 +1,56 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+"""Auto mixed precision."""
+import mindspore.nn as nn
+from mindspore.ops import functional as F
+from mindspore._checkparam import Validator as validator
+from mindspore.common import dtype as mstype
+
+
+class OutputTo(nn.Cell):
+    "Cast cell output back to float16 or float32"
+
+    def __init__(self, op, to_type=mstype.float16):
+        super(OutputTo, self).__init__(auto_prefix=False)
+        self._op = op
+        validator.check_type_name('to_type', to_type, [mstype.float16, mstype.float32], None)
+        self.to_type = to_type
+
+    def construct(self, x):
+        return F.cast(self._op(x), self.to_type)
+
+
+def auto_mixed_precision(network):
+    """Do keep batchnorm fp32."""
+    cells = network.name_cells()
+    change = False
+    network.to_float(mstype.float16)
+    for name in cells:
+        subcell = cells[name]
+        if subcell == network:
+            continue
+        elif name == 'fc':
+            network.insert_child_to_cell(name, OutputTo(subcell, mstype.float32))
+            change = True
+        elif name == 'conv2':
+            subcell.to_float(mstype.float32)
+            change = True
+        elif isinstance(subcell, (nn.BatchNorm2d, nn.BatchNorm1d)):
+            network.insert_child_to_cell(name, OutputTo(subcell.to_float(mstype.float32), mstype.float16))
+            change = True
+        else:
+            auto_mixed_precision(subcell)
+    if isinstance(network, nn.SequentialCell) and change:
+        network.cell_list = list(network.cells())

model_zoo/official/cv/resnext50/src/utils/cunstom_op.py

@@ -29,14 +29,10 @@ class GlobalAvgPooling(nn.Cell):
     """
     def __init__(self):
         super(GlobalAvgPooling, self).__init__()
-        self.mean = P.ReduceMean(True)
-        self.shape = P.Shape()
-        self.reshape = P.Reshape()
+        self.mean = P.ReduceMean(False)
 
     def construct(self, x):
         x = self.mean(x, (2, 3))
-        b, c, _, _ = self.shape(x)
-        x = self.reshape(x, (b, c))
         return x
 
 

model_zoo/official/cv/resnext50/train.py

@@ -36,11 +36,9 @@ from src.warmup_cosine_annealing_lr import warmup_cosine_annealing_lr
 from src.utils.logging import get_logger
 from src.utils.optimizers__init__ import get_param_groups
 from src.image_classification import get_network
+from src.utils.auto_mixed_precision import auto_mixed_precision
 from src.config import config
 
-devid = int(os.getenv('DEVICE_ID'))
-context.set_context(mode=context.GRAPH_MODE, enable_auto_mixed_precision=True,
-                    device_target="Ascend", save_graphs=False, device_id=devid)
 
 class BuildTrainNetwork(nn.Cell):
     """build training network"""
@@ -109,6 +107,7 @@ class ProgressMonitor(Callback):
 def parse_args(cloud_args=None):
     """parameters"""
     parser = argparse.ArgumentParser('mindspore classification training')
+    parser.add_argument('--platform', type=str, default='Ascend', choices=('Ascend', 'GPU'), help='run platform')
 
     # dataset related
     parser.add_argument('--data_dir', type=str, default='', help='train data dir')
@@ -141,6 +140,7 @@ def parse_args(cloud_args=None):
     args.label_smooth = config.label_smooth
     args.label_smooth_factor = config.label_smooth_factor
     args.ckpt_interval = config.ckpt_interval
+    args.ckpt_save_max = config.ckpt_save_max
     args.ckpt_path = config.ckpt_path
     args.is_save_on_master = config.is_save_on_master
     args.rank = config.rank
@@ -166,12 +166,25 @@ def merge_args(args, cloud_args):
 def train(cloud_args=None):
     """training process"""
     args = parse_args(cloud_args)
+    context.set_context(mode=context.GRAPH_MODE, enable_auto_mixed_precision=True,
+                        device_target=args.platform, save_graphs=False)
+    if os.getenv('DEVICE_ID', "not_set").isdigit():
+        context.set_context(device_id=int(os.getenv('DEVICE_ID')))
 
     # init distributed
     if args.is_distributed:
-        init()
+        if args.platform == "Ascend":
+            init()
+        else:
+            init("nccl")
         args.rank = get_rank()
         args.group_size = get_group_size()
+        parallel_mode = ParallelMode.DATA_PARALLEL
+        context.set_auto_parallel_context(parallel_mode=parallel_mode, device_num=args.group_size,
+                                          parameter_broadcast=True, mirror_mean=True)
+    else:
+        args.rank = 0
+        args.group_size = 1
 
     if args.is_dynamic_loss_scale == 1:
         args.loss_scale = 1  # for dynamic loss scale can not set loss scale in momentum opt
@@ -192,7 +205,7 @@ def train(cloud_args=None):
     # dataloader
     de_dataset = classification_dataset(args.data_dir, args.image_size,
                                         args.per_batch_size, 1,
-                                        args.rank, args.group_size)
+                                        args.rank, args.group_size, num_parallel_workers=8)
     de_dataset.map_model = 4  # !!!important
     args.steps_per_epoch = de_dataset.get_dataset_size()
 
@@ -201,15 +214,9 @@ def train(cloud_args=None):
     # network
     args.logger.important_info('start create network')
     # get network and init
-    network = get_network(args.backbone, args.num_classes)
+    network = get_network(args.backbone, args.num_classes, platform=args.platform)
     if network is None:
         raise NotImplementedError('not implement {}'.format(args.backbone))
-    network.add_flags_recursive(fp16=True)
-    # loss
-    if not args.label_smooth:
-        args.label_smooth_factor = 0.0
-    criterion = CrossEntropy(smooth_factor=args.label_smooth_factor,
-                             num_classes=args.num_classes)
 
     # load pretrain model
     if os.path.isfile(args.pretrained):
@@ -252,31 +259,29 @@ def train(cloud_args=None):
                    loss_scale=args.loss_scale)
 
 
-    criterion.add_flags_recursive(fp32=True)
+    # loss
+    if not args.label_smooth:
+        args.label_smooth_factor = 0.0
+    loss = CrossEntropy(smooth_factor=args.label_smooth_factor, num_classes=args.num_classes)
 
-    # package training process, adjust lr + forward + backward + optimizer
-    train_net = BuildTrainNetwork(network, criterion)
-    if args.is_distributed:
-        parallel_mode = ParallelMode.DATA_PARALLEL
-    else:
-        parallel_mode = ParallelMode.STAND_ALONE
     if args.is_dynamic_loss_scale == 1:
         loss_scale_manager = DynamicLossScaleManager(init_loss_scale=65536, scale_factor=2, scale_window=2000)
     else:
         loss_scale_manager = FixedLossScaleManager(args.loss_scale, drop_overflow_update=False)
 
-    # Model api changed since TR5_branch 2020/03/09
-    context.set_auto_parallel_context(parallel_mode=parallel_mode, device_num=args.group_size,
-                                      parameter_broadcast=True, mirror_mean=True)
-    model = Model(train_net, optimizer=opt, metrics=None, loss_scale_manager=loss_scale_manager)
+    if args.platform == "Ascend":
+        model = Model(network, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale_manager,
+                      metrics={'acc'}, amp_level="O3")
+    else:
+        auto_mixed_precision(network)
+        model = Model(network, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale_manager, metrics={'acc'})
 
     # checkpoint save
     progress_cb = ProgressMonitor(args)
     callbacks = [progress_cb,]
     if args.rank_save_ckpt_flag:
-        ckpt_max_num = args.max_epoch * args.steps_per_epoch // args.ckpt_interval
-        ckpt_config = CheckpointConfig(save_checkpoint_steps=args.ckpt_interval,
-                                       keep_checkpoint_max=ckpt_max_num)
+        ckpt_config = CheckpointConfig(save_checkpoint_steps=args.ckpt_interval * args.steps_per_epoch,
+                                       keep_checkpoint_max=args.ckpt_save_max)
         ckpt_cb = ModelCheckpoint(config=ckpt_config,
                                   directory=args.outputs_dir,
                                   prefix='{}'.format(args.rank))