merge commits fixing bugs from branch main to branch v0.4.0 (#799)

* fix: fix all zero initialized tensor problem for vit (#794)

Co-authored-by: ChongWei905 <weichong4@huawei.com>
(cherry picked from commit 2a07185)

* feat: add validate shuffle config and set default to False (#793)

(cherry picked from commit 1c283bc)

Co-authored-by: ChongWei905 <weichong4@huawei.com>
(cherry picked from commit 69ae862)

* fix: fix load checkpoint failure for deeplabv3 (#790)

Co-authored-by: ChongWei905 <weichong4@huawei.com>
(cherry picked from commit c744121)

* ops.ResizeBilinear has been deprecated, thus using ops.ResizeBilinearV2 instead (#789)

(cherry picked from commit a6d45f3)

---------

Co-authored-by: Pingqi Li <58093835+PingqiLi@users.noreply.github.com>

config.py

@@ -263,6 +263,8 @@ def create_parser():
                        help='Loss scale (default=1.0)')
     group.add_argument('--drop_overflow_update', type=bool, default=False,
                        help='Whether to execute optimizer if there is an overflow (default=False)')
+    group.add_argument('--eval_shuffle', type=bool, default=False,
+                       help='Whether to shuffle the evaluation data (default=False)')
 
     return parser_config, parser
 # fmt: on

examples/det/ssd/model.py

@@ -208,7 +208,7 @@ def construct(self, inputs):
             top = len(inputs) - i - 1
             down = top - 1
             size = ops.shape(inputs[down])
-            top_down = ops.ResizeBilinear((size[2], size[3]))(features[-1])
+            top_down = ops.ResizeBilinearV2()(features[-1], (size[2], size[3]))
             top_down = top_down + image_features[down]
             features = features + (top_down,)
 

examples/seg/deeplabv3/callbacks.py

@@ -100,7 +100,7 @@ def on_train_end(self, run_context):
 
 def get_segment_train_callback(args, steps_per_epoch, rank_id):
     callbacks = [TimeMonitor(data_size=steps_per_epoch), LossMonitor()]
-    if rank_id == 0:
+    if rank_id == 0 or rank_id is None:
         ckpt_config = CheckpointConfig(
             save_checkpoint_steps=args.save_steps,
             keep_checkpoint_max=args.keep_checkpoint_max,

examples/seg/deeplabv3/deeplabv3.py

@@ -260,7 +260,7 @@ class DeepLabInferNetwork(nn.Cell):
     """
 
     def __init__(self, network, input_format="NCHW"):
-        super(DeepLabInferNetwork, self).__init__()
+        super(DeepLabInferNetwork, self).__init__(auto_prefix=False)
         self.network = network
         self.softmax = nn.Softmax(axis=1)
         self.format = input_format

examples/seg/deeplabv3/train.py

@@ -140,7 +140,7 @@ def train(args):
     callbacks = get_segment_train_callback(args, steps_per_epoch, rank_id)
 
     # eval when train
-    if args.eval_while_train and rank_id == 0:
+    if args.eval_while_train and (rank_id == 0 or rank_id is None):
         eval_model = DeepLabInferNetwork(deeplab, input_format=args.input_format)
         eval_dataset = create_segment_dataset(
             name=args.dataset,

mindcv/models/vit.py

@@ -329,7 +329,9 @@ def get_num_layers(self):
     def _init_weights(self):
         w = self.patch_embed.proj.weight
         w_shape_flatted = (w.shape[0], functools.reduce(lambda x, y: x*y, w.shape[1:]))
-        w.set_data(initializer(XavierUniform(), w_shape_flatted, w.dtype).reshape(w.shape))
+        w_value = initializer(XavierUniform(), w_shape_flatted, w.dtype)
+        w_value.init_data()
+        w.set_data(w_value.reshape(w.shape))
         for _, cell in self.cells_and_names():
             if isinstance(cell, nn.Dense):
                 cell.weight.set_data(

validate.py

@@ -37,6 +37,7 @@ def validate(args):
         split=args.val_split,
         num_parallel_workers=args.num_parallel_workers,
         download=args.dataset_download,
+        shuffle=args.eval_shuffle,
     )
 
     # create transform