!6237 mobilenetv2 debug after test

Merge pull request !6237 from yepei6/modify_mobilenetv2

model_zoo/official/cv/mobilenetv2/README.md

@@ -4,17 +4,16 @@
 - [Model Architecture](#model-architecture)
 - [Dataset](#dataset)
 - [Features](#features)
-  - [Mixed Precision](#mixed-precision)
+  - [Mixed Precision](#mixed-precision(ascend))
 - [Environment Requirements](#environment-requirements)
 - [Script Description](#script-description)
   - [Script and Sample Code](#script-and-sample-code)
     - [Training Process](#training-process)
-    - [Evaluation Process](#evaluation-process)
+    - [Evaluation Process](#eval-process)
-      - [Evaluation](#evaluation)
 - [Model Description](#model-description)
   - [Performance](#performance)  
-    - [Training Performance](#evaluation-performance)
+    - [Training Performance](#training-performance)
-    - [Inference Performance](#evaluation-performance)
+    - [Evaluation Performance](#evaluation-performance)
 - [Description of Random Situation](#description-of-random-situation)
 - [ModelZoo Homepage](#modelzoo-homepage)
 
@@ -92,84 +91,84 @@ You can start training using python or shell scripts. The usage of shell scripts
 
 ### Launch
 
-```
+```shell
 # training example
   python:
-      Ascend: python train.py --dataset_path ~/imagenet/train/ --platform Ascend --train_method train
+      Ascend: python train.py --platform Ascend --dataset_path [TRAIN_DATASET_PATH] --train_method train
-      GPU: python train.py --dataset_path ~/imagenet/train/ --platform GPU --train_method train
+      GPU: python train.py --platform GPU --dataset_path [TRAIN_DATASET_PATH] --train_method train
-      CPU: python train.py --dataset_path ~/imagenet/train/ --platform CPU --train_method train
+      CPU: python train.py --platform CPU --dataset_path [TRAIN_DATASET_PATH] --train_method train
 
   shell:
-      Ascend: sh run_train.sh Ascend 8 0,1,2,3,4,5,6,7 hccl_config.json ~/imagenet/train/  train
+      Ascend: sh run_train.sh Ascend 8 0,1,2,3,4,5,6,7 hccl_config.json [TRAIN_DATASET_PATH]  train
-      GPU: sh run_train.sh GPU 8 0,1,2,3,4,5,6,7 ~/imagenet/train/ train
+      GPU: sh run_train.sh GPU 8 0,1,2,3,4,5,6,7 [TRAIN_DATASET_PATH] train
-      CPU: sh run_train.sh CPU ~/imagenet/train/ train
+      CPU: sh run_train.sh CPU [TRAIN_DATASET_PATH] train
 
 # fine tune example
   python:
-      Ascend: python train.py --dataset_path ~/imagenet/train/ --platform Ascend --train_method fine_tune ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      Ascend: python train.py --platform Ascend --dataset_path [TRAIN_DATASET_PATH] --train_method fine_tune ./pretrain_checkpoint/mobilenetv2.ckpt
-      GPU: python train.py --dataset_path ~/imagenet/train/ --platform GPU --train_method fine_tune ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      GPU: python train.py --platform GPU --dataset_path [TRAIN_DATASET_PATH] --train_method fine_tune ./pretrain_checkpoint/mobilenetv2.ckpt
-      CPU: python train.py --dataset_path ~/imagenet/train/ --platform CPU --train_method fine_tune ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      CPU: python train.py --platform CPU --dataset_path [TRAIN_DATASET_PATH] --train_method fine_tune ./pretrain_checkpoint/mobilenetv2.ckpt
 
   shell:
-      Ascend: sh run_train.sh Ascend 8 0,1,2,3,4,5,6,7 hccl_config.json ~/imagenet/train/  fine_tune ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      Ascend: sh run_train.sh Ascend 8 0,1,2,3,4,5,6,7 hccl_config.json [TRAIN_DATASET_PATH]  fine_tune ./pretrain_checkpoint/mobilenetv2.ckpt
-      GPU: sh run_train.sh GPU 8 0,1,2,3,4,5,6,7 ~/imagenet/train/ fine_tune ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      GPU: sh run_train.sh GPU 8 0,1,2,3,4,5,6,7 [TRAIN_DATASET_PATH] fine_tune ./pretrain_checkpoint/mobilenetv2.ckpt
-      CPU: sh run_train.sh CPU ~/imagenet/train/ fine_tune ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      CPU: sh run_train.sh CPU [TRAIN_DATASET_PATH] fine_tune ./pretrain_checkpoint/mobilenetv2.ckpt
 
 # incremental learn example
   python:
-      Ascend: python train.py --dataset_path ~/imagenet/train/ --platform Ascend --train_method incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      Ascend: python --platform Ascend train.py --dataset_path [TRAIN_DATASET_PATH] --train_method incremental_learn ./pretrain_checkpoint/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      GPU: python train.py --dataset_path ~/imagenet/train/ --platform GPU --train_method incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      GPU: python --platform GPU train.py --dataset_path [TRAIN_DATASET_PATH] --train_method incremental_learn ./pretrain_checkpoint/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      CPU: python train.py --dataset_path ~/imagenet/train/ --platform CPU --train_method incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      CPU: python --platform CPU train.py --dataset_path [TRAIN_DATASET_PATH] --train_method incremental_learn ./pretrain_checkpoint/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
 
   shell:
-      Ascend: sh run_train.sh Ascend 8 0,1,2,3,4,5,6,7 hccl_config.json ~/imagenet/train/  incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      Ascend: sh run_train.sh Ascend 8 0,1,2,3,4,5,6,7 hccl_config.json [TRAIN_DATASET_PATH]  incremental_learn ./pretrain_checkpoint/mobilenetv2.ckpt  ./checkpoint/mobilenetv2_head_15.ckpt
-      GPU: sh run_train.sh GPU 8 0,1,2,3,4,5,6,7 ~/imagenet/train/ incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      GPU: sh run_train.sh GPU 8 0,1,2,3,4,5,6,7 [TRAIN_DATASET_PATH] incremental_learn ./pretrain_checkpoint/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      CPU: sh run_train.sh CPU ~/imagenet/train/ incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt
+      CPU: sh run_train.sh CPU [TRAIN_DATASET_PATH] incremental_learn ./pretrain_checkpoint/mobilenetv2_199.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
 ```
 
 ### Result
 
-Training result will be stored in the example path. Checkpoints will be stored at `. /checkpoint` by default, and training log  will be redirected to `./train/train.log` like followings.
+Training result will be stored in the example path. Checkpoints will be stored at `. /checkpoint` by default, and training log  will be redirected to `./train.log` like followings with the platform CPU and GPU, will be wrote to `./train/rank*/log*.log` with the platform Ascend .
 
-```
+```shell
 epoch: [  0/200], step:[  624/  625], loss:[5.258/5.258], time:[140412.236], lr:[0.100]
 epoch time: 140522.500, per step time: 224.836, avg loss: 5.258
 epoch: [  1/200], step:[  624/  625], loss:[3.917/3.917], time:[138221.250], lr:[0.200]
 epoch time: 138331.250, per step time: 221.330, avg loss: 3.917
 ```
 
-## [Eval process](#contents)
+## [Evaluation process](#contents)
 
 ### Usage
 
-You can start training using python or shell scripts. The usage of shell scripts as follows:
+You can start training using python or shell scripts.If the train method is train or fine tune, should not input the `[CHECKPOINT_PATH]` The usage of shell scripts as follows:
 
-- Ascend: sh run_infer.sh Ascend [DATASET_PATH] [CHECKPOINT_PATH] [HEAD_CKPT_PATH]
+- Ascend: sh run_eval.sh Ascend [DATASET_PATH] [CHECKPOINT_PATH] [HEAD_CKPT_PATH]
-- GPU: sh run_infer.sh GPU [DATASET_PATH] [CHECKPOINT_PATH] [HEAD_CKPT_PATH]
+- GPU: sh run_eval.sh GPU [DATASET_PATH] [CHECKPOINT_PATH] [HEAD_CKPT_PATH]
-- CPU: sh run_infer.sh CPU [DATASET_PATH] [BACKBONE_CKPT_PATH] [HEAD_CKPT_PATH]
+- CPU: sh run_eval.sh CPU [DATASET_PATH] [BACKBONE_CKPT_PATH] [HEAD_CKPT_PATH]
 
 ### Launch
 
-```
+```shell
-# infer example
+# eval example
   python:
-      Ascend: python eval.py --dataset_path ~/imagenet/val/ --pretrain_ckpt ~/train/mobilenet-200_625.ckpt --platform Ascend --head_ckpt ./checkpoint/mobilenetv2_199.ckpt
+      Ascend: python eval.py --platform Ascend --dataset_path [VAL_DATASET_PATH] --pretrain_ckpt ./pretrain_ckpt/mobilenetv2.ckpt --head_ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      GPU: python eval.py --dataset_path ~/imagenet/val/ --pretrain_ckpt ~/train/mobilenet-200_625.ckpt --platform GPU --head_ckpt ./checkpoint/mobilenetv2_199.ckpt
+      GPU: python eval.py --platform GPU --dataset_path [VAL_DATASET_PATH] --pretrain_ckpt ./pretrain_ckpt/mobilenetv2.ckpt --head_ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      CPU: python eval.py --dataset_path ~/imagenet/val/ --pretrain_ckpt ~/train/mobilenet-200_625.ckpt --platform CPU --head_ckpt ./checkpoint/mobilenetv2_199.ckpt
+      CPU: python eval.py --platform CPU --dataset_path [VAL_DATASET_PATH] --pretrain_ckpt ./pretrain_ckpt/mobilenetv2.ckpt --head_ckpt ./checkpoint/mobilenetv2_head_15.ckpt
 
   shell:
-      Ascend: sh run_infer.sh Ascend ~/imagenet/val/ ~/train/mobilenet-200_625.ckpt ./checkpoint/mobilenetv2_199.ckpt
+      Ascend: sh run_eval.sh Ascend [VAL_DATASET_PATH] ./pretrain_ckpt/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      GPU: sh run_infer.sh GPU ~/imagenet/val/ ~/train/mobilenet-200_625.ckpt ./checkpoint/mobilenetv2_199.ckpt
+      GPU: sh run_eval.sh GPU [VAL_DATASET_PATH] ./pretrain_ckpt/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
-      CPU: sh run_infer.sh GPU ~/imagenet/val/ ~/train/mobilenet-200_625.ckpt ./checkpoint/mobilenetv2_199.ckpt
+      CPU: sh run_eval.sh CPU [VAL_DATASET_PATH] ./pretrain_ckpt/mobilenetv2.ckpt ./checkpoint/mobilenetv2_head_15.ckpt
 ```
 
 > checkpoint can be produced in training process.
 
 ### Result
 
-Inference result will be stored in the example path, you can find result like the followings in `val.log`.
+Inference result will be stored in the example path, you can find result like the followings in `eval.log`.
 
-```
+```shell
 result: {'acc': 0.71976314102564111} ckpt=/path/to/checkpoint/mobilenet-200_625.ckpt
 ```
 
@@ -199,7 +198,8 @@ result: {'acc': 0.71976314102564111} ckpt=/path/to/checkpoint/mobilenet-200_625.
 
 # [Description of Random Situation](#contents)
 
-In dataset.py, we set the seed inside “create_dataset" function. We also use random seed in train.py.
+<!-- In dataset.py, we set the seed inside “create_dataset" function. We also use random seed in train.py. -->
+In train.py, we set the seed which is used by numpy.random, mindspore.common.Initializer, mindspore.ops.composite.random_ops and mindspore.nn.probability.distribution.
 
 # [ModelZoo Homepage](#contents)
 

model_zoo/official/cv/mobilenetv2/scripts/run_eval.sh

@@ -18,10 +18,10 @@
 
 run_ascend()
 {
-    # check checkpoint file
+    # check pretrain_ckpt file
     if [ ! -f $3 ]
     then
-        echo "error: CHECKPOINT_PATH=$3 is not a file"
+        echo "error: PRETRAIN_CKPT=$3 is not a file"
     exit 1
     fi
 
@@ -44,15 +44,15 @@ run_ascend()
             --dataset_path=$2 \
             --pretrain_ckpt=$3 \
             --head_ckpt=$4 \
-            &> ../infer.log &  # dataset val folder path
+            &> ../eval.log &  # dataset val folder path
 }
 
 run_gpu()
 {
-    # check checkpoint file
+    # check pretrain_ckpt file
     if [ ! -f $3 ]
     then
-        echo "error: CHECKPOINT_PATH=$3 is not a file"
+        echo "error: PRETRAIN_CKPT=$3 is not a file"
     exit 1
     fi
 
@@ -70,26 +70,18 @@ run_gpu()
         --dataset_path=$2 \
         --pretrain_ckpt=$3 \
         --head_ckpt=$4 \
-        &> ../infer.log &  # dataset train folder
+        &> ../eval.log &  # dataset train folder
 }
 
 run_cpu()
 {
-    # check checkpoint file
+    # check pretrain_ckpt file
     if [ ! -f $3 ]
     then
-        echo "error: BACKBONE_CKPT=$3 is not a file"
+        echo "error: PRETRAIN_CKPT=$3 is not a file"
     exit 1
     fi
 
-    # check checkpoint file
-    if [ ! -f $4 ]
-    then
-        echo "error: HEAD_CKPT=$4 is not a file"
-    exit 1
-    fi
-
-
     BASEPATH=$(cd "`dirname $0`" || exit; pwd)
     export PYTHONPATH=${BASEPATH}:$PYTHONPATH
     if [ -d "../eval" ];
@@ -104,13 +96,14 @@ run_cpu()
         --dataset_path=$2 \
         --pretrain_ckpt=$3 \
         --head_ckpt=$4 \
-        &> ../infer.log &  # dataset train folder
+        &> ../eval.log &  # dataset train folder
 }
 
 
 if [ $# -gt 4 ] || [ $# -lt 3 ]
 then
-    echo "Ascend: sh run_infer.sh [PLATFORM] [DATASET_PATH] [PRETRAIN_CKPT] \
+    echo "Usage:
+          Ascend: sh run_infer.sh [PLATFORM] [DATASET_PATH] [PRETRAIN_CKPT]
           GPU: sh run_infer.sh [PLATFORM] [DATASET_PATH] [PRETRAIN_CKPT]
           CPU: sh run_infer.sh [PLATFORM] [DATASET_PATH] [BACKBONE_CKPT] [HEAD_CKPT]"
 exit 1

model_zoo/official/cv/mobilenetv2/scripts/run_train.sh

@@ -109,11 +109,10 @@ run_cpu()
 
 if [ $# -gt 7 ] || [ $# -lt 4 ]
 then
-    echo "Usage:\n \
+    echo "Usage:
-          Ascend: sh run_train.sh Ascend [DEVICE_NUM] [VISIABLE_DEVICES(0,1,2,3,4,5,6,7)] [RANK_TABLE_FILE] [DATASET_PATH] [TRAIN_METHOD] [CKPT_PATH] \n \
+          Ascend: sh run_train.sh Ascend [DEVICE_NUM] [VISIABLE_DEVICES(0,1,2,3,4,5,6,7)] [RANK_TABLE_FILE] [DATASET_PATH] [TRAIN_METHOD] [CKPT_PATH]
-          GPU: sh run_train.sh GPU [DEVICE_NUM] [VISIABLE_DEVICES(0,1,2,3,4,5,6,7)] [DATASET_PATH] [TRAIN_METHOD] [CKPT_PATH]\n \
+          GPU: sh run_train.sh GPU [DEVICE_NUM] [VISIABLE_DEVICES(0,1,2,3,4,5,6,7)] [DATASET_PATH] [TRAIN_METHOD] [CKPT_PATH]
-          CPU: sh run_train.sh CPU [DATASET_PATH] [TRAIN_METHOD] [CKPT_PATH]\n \
+          CPU: sh run_train.sh CPU [DATASET_PATH] [TRAIN_METHOD] [CKPT_PATH]"
-          "
 exit 1
 fi
 

model_zoo/official/cv/mobilenetv2/src/args.py

@@ -38,31 +38,28 @@ def launch_parse_args():
 
 def train_parse_args():
     train_parser = argparse.ArgumentParser(description='Image classification trian')
-    train_parser.add_argument('--dataset_path', type=str, required=True, help='Dataset path')
     train_parser.add_argument('--platform', type=str, default="Ascend", choices=("CPU", "GPU", "Ascend"), \
         help='run platform, only support CPU, GPU and Ascend')
-    train_parser.add_argument('--pretrain_ckpt', type=str, default=None, help='Pretrained checkpoint path \
+    train_parser.add_argument('--dataset_path', type=str, required=True, help='Dataset path')
-        for fine tune or incremental learning')
-    train_parser.add_argument('--run_distribute', type=ast.literal_eval, default=True, help='Run distribute')
     train_parser.add_argument('--train_method', type=str, choices=("train", "fine_tune", "incremental_learn"), \
         help="\"fine_tune\"or \"incremental_learn\" if to fine tune the net  after loading the ckpt, \"train\" to \
         train from initialization model")
-
+    train_parser.add_argument('--pretrain_ckpt', type=str, default=None, help='Pretrained checkpoint path \
+        for fine tune or incremental learning')
+    train_parser.add_argument('--run_distribute', type=ast.literal_eval, default=True, help='Run distribute')
     train_args = train_parser.parse_args()
     return train_args
 
 def eval_parse_args():
     eval_parser = argparse.ArgumentParser(description='Image classification eval')
-    eval_parser.add_argument('--dataset_path', type=str, required=True, help='Dataset path')
     eval_parser.add_argument('--platform', type=str, default="Ascend", choices=("Ascend", "GPU", "CPU"), \
         help='run platform, only support GPU, CPU and Ascend')
-    eval_parser.add_argument('--pretrain_ckpt', type=str, default=None, help='Pretrained checkpoint path \
+    eval_parser.add_argument('--dataset_path', type=str, required=True, help='Dataset path')
+    eval_parser.add_argument('--pretrain_ckpt', type=str, required=True, help='Pretrained checkpoint path \
         for fine tune or incremental learning')
     eval_parser.add_argument('--head_ckpt', type=str, default=None, help='Pretrained checkpoint path \
-        for fine tune or incremental learning')
+        for incremental learning')
-    eval_parser.add_argument('--run_distribute', type=ast.literal_eval, default=False, help='Run distribute')
+    eval_parser.add_argument('--run_distribute', type=ast.literal_eval, default=False, help='If run distribute in GPU.')
-
     eval_args = eval_parser.parse_args()
-
     return eval_args
     

model_zoo/official/cv/mobilenetv2/src/dataset.py

@@ -16,7 +16,6 @@
 create train or eval dataset.
 """
 import os
-from tqdm import tqdm
 import numpy as np
 
 from mindspore import Tensor
@@ -97,7 +96,11 @@ def create_dataset(dataset_path, do_train, config, repeat_num=1):
     # apply dataset repeat operation
     ds = ds.repeat(repeat_num)
 
-    return ds
+    step_size = ds.get_dataset_size()
+    if step_size == 0:
+        raise ValueError("The step_size of dataset is zero. Check if the images of train dataset is more than batch_\
+            size in config.py")
+    return ds, step_size
 
 
 def extract_features(net, dataset_path, config):
@@ -109,19 +112,16 @@ def extract_features(net, dataset_path, config):
                              config=config,
                              repeat_num=1)
     step_size = dataset.get_dataset_size()
-    pbar = tqdm(list(dataset.create_dict_iterator(output_numpy=True)))
     model = Model(net)
-    i = 0
+
-    for data in pbar:
+    for i, data in enumerate(dataset.create_dict_iterator(output_numpy=True)):
         features_path = os.path.join(features_folder, f"feature_{i}.npy")
         label_path = os.path.join(features_folder, f"label_{i}.npy")
-        if not (os.path.exists(features_path) and os.path.exists(label_path)):
+        if not os.path.exists(features_path or not os.path.exists(label_path)):
             image = data["image"]
             label = data["label"]
             features = model.predict(Tensor(image))
             np.save(features_path, features.asnumpy())
             np.save(label_path, label)
-        pbar.set_description("Process dataset batch: %d" % (i + 1))
+        print(f"Complete the batch {i}/{step_size}")
-        i += 1
-
     return step_size

model_zoo/official/cv/mobilenetv2/src/launch.py

@@ -38,17 +38,17 @@ def main():
     for rank_id in range(0, args.nproc_per_node):
         os.chdir(cur_path)
         device_id = visible_devices[rank_id]
-        device_dir = os.path.join(cur_path, 'device{}'.format(rank_id))
+        rank_dir = os.path.join(cur_path, 'rank{}'.format(rank_id))
         env['RANK_ID'] = str(rank_id)
         env['DEVICE_ID'] = str(device_id)
-        if os.path.exists(device_dir):
+        if os.path.exists(rank_dir):
-            shutil.rmtree(device_dir)
+            shutil.rmtree(rank_dir)
-        os.mkdir(device_dir)
+        os.mkdir(rank_dir)
-        os.chdir(device_dir)
+        os.chdir(rank_dir)
         cmd = [sys.executable, '-u']
         cmd.append(args.training_script)
         cmd.extend(args.training_script_args)
-        log_file = open('{dir}/log{id}.log'.format(dir=device_dir, id=rank_id), 'w')
+        log_file = open(f'{rank_dir}/log{rank_id}.log', 'w')
         process = subprocess.Popen(cmd, stdout=log_file, stderr=log_file, env=env)
         processes.append(process)
         cmds.append(cmd)

model_zoo/official/cv/mobilenetv2/train.py

@@ -60,8 +60,7 @@ if __name__ == '__main__':
     elif args_opt.train_method in ("train", "fine_tune"):
         if args_opt.platform == "CPU":
             raise ValueError("Currently, CPU only support \"incremental_learn\", not \"fine_tune\" or \"train\".")
-        dataset = create_dataset(dataset_path=args_opt.dataset_path, do_train=True, config=config)
+        dataset, step_size = create_dataset(dataset_path=args_opt.dataset_path, do_train=True, config=config)
-        step_size = dataset.get_dataset_size()
 
     # Currently, only Ascend support switch precision.
     switch_precision(net, mstype.float16, config)
@@ -108,9 +107,8 @@ if __name__ == '__main__':
                 losses.append(network(feature, label).asnumpy())
             epoch_mseconds = (time.time()-epoch_start) * 1000
             per_step_mseconds = epoch_mseconds / step_size
-            print("\r epoch[{}], iter[{}] cost: {:5.3f}, per step time: {:5.3f}, avg loss: {:5.3f}"\
+            print("epoch[{}], iter[{}] cost: {:5.3f}, per step time: {:5.3f}, avg loss: {:5.3f}"\
-            .format(epoch + 1, step_size, epoch_mseconds, per_step_mseconds, np.mean(np.array(losses))), \
+            .format(epoch + 1, step_size, epoch_mseconds, per_step_mseconds, np.mean(np.array(losses))))
-                end="")
             if (epoch + 1) % config.save_checkpoint_epochs == 0:
                 save_checkpoint(network, os.path.join(config.save_checkpoint_path, \
                     f"mobilenetv2_head_{epoch+1}.ckpt"))