yankf21
/
mindspore2022
forked from huawei/mindspore2022
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
mindspore2022/model_zoo/research/cv/SE-Net/README.md

15 KiB
Raw Blame History

Contents

SE-Net Description

Description

something should be written here.

Paper

paper:Jie Hu, Li Shen, Samuel Albanie, Gang Sun, Enhua Wu. "Squeeze-and-Excitation Networks"

Model Architecture

The overall network architecture of Net is show below: Link

Dataset

Dataset used: ImageNet2012

  • Dataset size 224*224 colorful images in 1000 classes
    • Train：1,281,167 images
    • Test： 50,000 images
  • Data format：jpeg
    • Note：Data will be processed in dataset.py

Features

Mixed Precision

The mixed precision training method accelerates the deep learning neural network training process by using both the single-precision and half-precision data types, and maintains the network precision achieved by the single-precision training at the same time. Mixed precision training can accelerate the computation process, reduce memory usage, and enable a larger model or batch size to be trained on specific hardware. For FP16 operators, if the input data type is FP32, the backend of MindSpore will automatically handle it with reduced precision. Users could check the reduced-precision operators by enabling INFO log and then searching ‘reduce precision’.

Environment Requirements

Quick Start

After installing MindSpore via the official website, you can start training and evaluation as follows:

  • Running on Ascend
# distributed training
Usage:
bash run_distribute_train.sh se-resnet50 imagenet2012 [RANK_TABLE_FILE] [DATASET_PATH]

# standalone training
Usage:
export DEVICE_ID=0
python train.py --net=se-resnet50  --dataset=imagenet2012  --dataset_path=[DATASET_PATH]

# run evaluation example
Usage:
export DEVICE_ID=0
python eval.py --net=se-resnet50 --dataset=imagenet2012 --checkpoint_path=[CHECKPOINT_PATH] --dataset_path=[DATASET_PATH]
  • Running on GPU
# distributed training
Usage:
sh run_distribute_train_gpu.sh se-resnet50 imagenet2012 [DATASET_PATH]

# standalone training
Usage:
sh run_standalone_train_gpu.sh se-resnet50 imagenet2012 [DATASET_PATH]

# run evaluation example
Usage:
sh run_eval_gpu.sh [DATASET_PATH] [CHECKPOINT_PATH]

Script Description

Script and Sample Code

.
└──SE-Net
  ├── README.md
  ├── ascend310_infer                      # application for 310 inference
  ├── scripts
    ├── run_distribute_train.sh            # launch ascend distributed training(8 pcs)
    ├── run_distribute_train_gpu.sh        # launch gpu distributed training(8 pcs)
    ├── run_eval.sh                        # launch ascend evaluation
    ├── run_eval_gpu.sh                    # launch gpu evaluation
    ├── run_standalone_train.sh            # launch ascend standalone training(1 pcs)
    ├── run_standalone_train_gpu.sh        # launch gpu standalone training(1 pcs)
    └─ run_infer_310.sh                    # shell script for 310inference on ascend
  ├── src
    ├── config.py                          # parameter configuration
    ├── CrossEntropySmooth.py              # loss definition for ImageNet2012 dataset
    ├── dataset.py                         # data preprocessing
    ├── lr_generator.py                    # generate learning rate for each step
    ├── resnet.py                          # resnet50 backbone
    └── se.py                              # se-block definition
  ├── export.py                            # export model for inference
  ├── eval.py                              # eval net
  ├── postprocess.py                       # postprocess scripts
  └── train.py                             # train net

Script Parameters

Parameters for both training and evaluation can be set in config.py.

  • Config for SE-ResNet50, ImageNet2012 dataset
"class_num": 1001,                # dataset class number
"batch_size": 256,                 # batch size of input tensor
"loss_scale": 1024,               # loss scale
"momentum": 0.9,                  # momentum optimizer
"weight_decay": 1e-4,             # weight decay
"epoch_size": 90,                 # only valid for taining, which is always 1 for inference
"pretrain_epoch_size": 0,         # epoch size that model has been trained before loading pretrained checkpoint, actual training epoch size is equal to epoch_size minus pretrain_epoch_size
"save_checkpoint": True,          # whether save checkpoint or not
"save_checkpoint_epochs": 5,      # the epoch interval between two checkpoints. By default, the last checkpoint will be saved after the last epoch
"keep_checkpoint_max": 10,        # only keep the last keep_checkpoint_max checkpoint
"save_checkpoint_path": "./",     # path to save checkpoint relative to the executed path
"warmup_epochs": 0,               # number of warmup epoch
"lr_decay_mode": "linear",        # decay mode for generating learning rate
"use_label_smooth": True,         # label smooth
"label_smooth_factor": 0.1,       # label smooth factor
"lr_init": 0,                     # initial learning rate
"lr_max": 0.8,                    # maximum learning rate
"lr_end": 0.0,                    # minimum learning rate

Training Process

Usage

Running on Ascend

# distributed training
Usage:
bash run_distribute_train.sh se-resnet50 imagenet2012  /imagenet/train  /rank_table.json

# standalone training
Usage:
export DEVICE_ID=0
bash run_standalone_train.sh  se-resnet50  imagenet2012   /data/imagenet/train/

Running on GPU

# distributed training
Usage:
sh run_distribute_train_gpu.sh se-resnet50 imagenet2012 [DATASET_PATH]

# standalone training
Usage:
sh run_standalone_train_gpu.sh se-resnet50 imagenet2012 [DATASET_PATH]

For distributed training, a hccl configuration file with JSON format needs to be created in advance.

Please follow the instructions in the link hccn_tools.

Training result will be stored in the example path, whose folder name begins with "train" or "train_parallel". Under this, you can find checkpoint file together with result like the following in log.

Result

  • Training SE-ResNet50 with ImageNet2012 dataset
# distribute training result(8 pcs)
epoch: 1 step: 625, loss is 5.0938025
epoch time: 303139.271 ms, per step time: 485.023 ms
epoch: 2 step: 625, loss is 4.152817
epoch time: 205321.853 ms, per step time: 328.515 ms
epoch: 3 step: 625, loss is 3.7530446
epoch time: 205214.637 ms, per step time: 328.343 ms
...
epoch: 89 step: 625, loss is 1.9109731
epoch time: 205217.996 ms, per step time: 328.349 ms
epoch: 90 step: 625, loss is 1.5931969
epoch time: 206295.838 ms, per step time: 330.073 ms

Evaluation Process

Usage

Running on Ascend

export DEVICE_ID=0
bash run_eval.sh /imagenet/val/  /path/to/resnet-90_625.ckpt

Running on GPU

sh run_eval_gpu.sh [DATASET_PATH] [CHECKPOINT_PATH]

Result

  • Evaluating SE-ResNet50 with ImageNet2012 dataset
result: {'top_5_accuracy': 0.9385269007731959, 'top_1_accuracy': 0.7774645618556701}

Inference Process

Export MindIR

python export.py --ckpt_file [CKPT_PATH] --batch_size [BATCH_SIZE] --file_format [FILE_FORMAT]

The ckpt_file parameter is required, FILE_FORMAT should be in ["AIR", "MINDIR"] BATCH_SIZE current batch_size can only be set to 1.

Infer on Ascend310

Before performing inference, the mindir file must be exported by export.py script. We only provide an example of inference using MINDIR model.

# Ascend310 inference
bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DVPP] [DEVICE_ID]
  • DVPP is mandatory, and must choose from ["DVPP", "CPU"], it's case-insensitive. SE-net only support CPU mode.
  • DEVICE_ID is optional, default value is 0.

result

Inference result is saved in current path, you can find result like this in acc.log file.

result: {'top_5_accuracy': 93.86%, 'top_1_accuracy': 77.80%}

Model Description

Performance

Evaluation Performance

SE-ResNet50 on ImageNet2012

Parameters Ascend GPU
Model Version SE-ResNet50 SE-ResNet50
Resource CentOs 8.2, Ascend 910，CPU 2.60GHz 192cores，Memory 755G V100-PCIE 32G
uploaded Date 07/05/2021 (month/day/year) 07/14/2021 (month/day/year)
MindSpore Version 1.3.0 1.3.0
Dataset ImageNet2012 ImageNet2012
Training Parameters epoch=90, steps per epoch=5004, batch_size = 256 epoch=90, steps per epoch=5004, batch_size = 256
Optimizer Momentum Momentum
Loss Function Softmax Cross Entropy Softmax Cross Entropy
outputs probability probability
Loss 1.5931969 1.6664593
Speed # ms/step（8pcs） 8pcs: 1016.9 ms/step
Total time # mins 8pcs: 15.9 hours
Parameters (M) 285M 285M
Checkpoint for Fine tuning # M (.ckpt file) # M (.ckpt file)
Scripts https://gitee.com/mindspore/mindspore/tree/master/model_zoo/research/cv/SE-Net https://gitee.com/mindspore/mindspore/tree/master/model_zoo/research/cv/SE-Net

Inference Performance

SE-ResNet50 on ImageNet2012

Parameters Ascend GPU
Model Version SE-ResNet50 SE-ResNet50
Resource Ascend 910 V100-PCIE 32G
Uploaded Date 07/05/2021 (month/day/year) 07/14/2021 (month/day/year)
MindSpore Version 1.3.0 1.3.0
Dataset ImageNet2012 ImageNet2012
batch_size 256 256
Accuracy 77.74% 77.66%
Model for inference # (.air file) ------

310Inference Performance

SE-ResNet50 on ImageNet2012

Parameters Ascend
Model Version SE-ResNet50
Resource Ascend 310; OS Euler2.8
Uploaded Date 06/16/2021 (month/day/year)
MindSpore Version 1.2.0
Dataset ImageNet2012
batch_size 1
outputs Accuracy
Accuracy 77.80%
Model for inference 285M(.ckpt file)

Description of Random Situation

In dataset.py, we set the seed inside "create_dataset" function. We also use random seed in train.py.

ModelZoo Homepage

Please check the official homepage.