• Contents
• ResNetV2-50-FRN Description
• Model architecture
• Dataset
• Environment Requirements
• Script description
• • Script and sample code
  • Script Parameters
  • Training Process
  • • Usage
    • Launch
  • Evaluation Process
  • • Usage
    • Launch
    • Result
• Model description
• • Performance
  • • Training Performance
    • Inference Performance
• ModelZoo Homepage

Contents

• ResNetV2-50-FRN Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
  • Script and Sample Code
  • Script Parameters
  • Training Process
  • Evaluation Process
• Model Description
  • Performance
    • Training Performance
    • Inference Performance
• ModelZoo Homepage

ResNetV2-50-FRN Description

Paper: Saurabh Singh, Shankar Krishnan. Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks. 2020.

Model architecture

The overall network architecture of ResNetV2-50-FRN is show below:

Link

Dataset

Dataset used: imagenet

• Dataset size: ~125G, 1.2M colorful images in 1000 classes
  • Train: 120G, 1.2M images
  • Test: 5G, 50000 images
• Data format: RGB images.
  • Note: Data will be processed in src/dataset.py

Environment Requirements

• Hardware Ascend
  • Prepare hardware environment with Ascend processor.
• Framework
  • MindSpore
• For more information, please check the resources below：
  • MindSpore Tutorials
  • MindSpore Python API

Script description

Script and sample code

.
└─resnetv2_50_frn
  ├─README.md
  ├─scripts
    ├─run_standalone_train_for_ascend.sh   # launch standalone training with Ascend platform(1p)
    ├─run_distribute_train_for_ascend.sh   # launch distributed training with Ascend platform(8p)
    ├─run_standalone_train_for_gpu.sh      # launch standalone training with gpu platform(1p)
    ├─run_distribute_train_for_gpu.sh      # launch distributed training with gpu platform(8p)
    ├─run_eval_for_ascend                  # launch evaluating with Ascend platform
    └─run_eval_for_gpu.sh                  # launch evaluating with gpu platform
  ├─src
    ├─ model_utils
       ├──config.py                        # parameter configuration
       ├──device_adapter.py                # device adapter
       ├──local_adapter.py                 # local adapter
       └──moxing_adapter.py                # moxing adapter
    ├─dataset.py                           # data preprocessing
    ├─lr_generator.py                      # learning rate generator
    └─resnetv2_50_frn.py                   # network definition
  ├─default_config.yaml                    # parameter configuration
  ├─export.py                              # convert checkpoint
  ├─eval.py                                # eval net
  └─train.py                               # train net

Script Parameters

Parameters for both training and evaluating can be set in default_config.yaml.

'random_seed': 1,                          # fix random seed
'rank': 0,                                 # local rank of distributed
'group_size': 1,                           # world size of distributed
'work_nums': 8,                            # number of workers to read the data
'epoch_size': 240,                         # total epoch numbers
'keep_checkpoint_max': 20,                 # max numbers to keep checkpoints
'save_ckpt_path': './',                    # save checkpoint path
'train_batch_size': 32,                    # input batch size for training
'val_batch_size': 125,                     # input batch size for evaluation
'num_classes': 1000,                       # dataset class numbers
'lr_init': 0.025,                          # initiate learning rate
'weight_decay': 0.0001,                    # weight decay
'momentum': 0.9,                           # momentum
'cutout': True,                            # whether to cutout the input data for training
'coutout_leng': 56,                        # the length of cutout when cutout is True

Training Process

Usage

Ascend:
    # distribute training example(8p)
    bash run_distribute_train_for_ascend.sh DATA_DIR
    # standalone training
    bash run_standalone_train_for_ascend.sh DEVICE_ID DATA_DIR

Launch

# distributed training example(8p) for Ascend
bash scripts/run_distribute_train_for_ascend.sh /dataset
# standalone training example for for Ascend
bash scripts/run_standalone_train_for_ascend.sh 0 /dataset

You can find checkpoint file together with result in log.

Evaluation Process

Usage

# Evaluation
bash run_eval_for_ascend.sh DEVICE_ID DATA_DIR PATH_CHECKPOINT

Launch

# Evaluation with checkpoint
bash scripts/run_eval_for_ascend.sh 0 /dataset ./ckpt_0/resnetv2-50-frn-rank0-240_5005.ckpt

Result

Evaluation result will be stored in the scripts path. Under this, you can find result like the followings in log.

acc=77.4%(TOP1,size:224224)
acc=78.3%(TOP1,size:299299)

Model description

Performance

Training Performance

Parameters	Ascend 910
Model Version	ResetNetV2-50-FRN
Resource	Ascend 910
uploaded Date	10/20/2021 (month/day/year)
MindSpore Version	1.3.0
Dataset	ImageNet
Training Parameters	default_config.yaml
Optimizer	SGD
Loss Function	SoftmaxCrossEntropyWithLogits
Loss	0.9140
Total time	50.5h 8ps
Checkpoint for Fine tuning	311.9 M(.ckpt file)

Inference Performance

Parameters	Ascend 910
Model Version	ResetNetV2-50-FRN
Resource	Ascend 910
uploaded Date	10/20/2021 (month/day/year)
MindSpore Version	1.3.0
Dataset	ImageNet
batch_size	125
outputs	probability
Accuracy	acc=77.4% (TOP1,size:224*224)
Accuracy	acc=78.3%(TOP1,size:299*299)

ModelZoo Homepage

Please check the official homepage.