• Contents
• LLNet Description
• Model architecture
• Dataset
• Environment Requirements
• Script description
• • Script and sample code
  • Script Parameters
  • Training Process
  • • Usage
    • Launch
  • Evaluation Process
  • • Usage
    • Launch
    • Result
  • Inference Process
  • • Export MindIR
    • Infer on Ascend310
    • result
• Model description
• • Performance
  • • Training Performance
    • Inference Performance
• ModelZoo Homepage

Contents

• LLNet Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
  • Script and Sample Code
  • Script Parameters
  • Training Process
  • Evaluation Process
  • Inference Process
    • Export MindIR
    • Infer on Ascend310
    • result
• Model Description
  • Performance
    • Training Performance
    • Inference Performance
• ModelZoo Homepage

LLNet Description

The authors proposed a deep autoencoder-based approach to identify signal features from low-light images handcrafting and adaptively brighten images without over-amplifying the lighter parts in images (i.e., without saturation of image pixels) in high dynamic range. The network can then be successfully applied to naturally low-light environment and/or hardware degraded images.

Paper: Kin Gwn Lore, Adedotun Akintayo, Soumik Sarkar: LLNet: A Deep Autoencoder Approach to Natural Low-light Image Enhancement. 2015.

Model architecture

The overall description of LLNet is show below:

Link

Dataset

Dataset used: dbimagenes

• Dataset size: 1.1G, 170 gray images
  • Train: 526.3M, 163 images, 1250 patches per image.
  • Val: 526.3M, 163 images, 1250 patches per image.
  • Test: 1.3M, 5 images, 29241 patches per image.
• Data format: gray 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

.
└─llnet
  ├─ README.md
  ├─ README_CN.md
  ├─── ascend310_infer
  │    ├── build.sh                            # build bash
  │    ├── CMakeLists.txt                      # CMakeLists
  │    ├── inc
  │    │   └── utils.h                         # utils head
  │    └── src
  │        ├── main.cc                         # main function of ascend310_infer
  │        └── utils.cc                        # utils function of ascend310_infer
  ├─── scripts
  │    ├─ run_standalone_train_for_ascend.sh   # launch standalone training with Ascend 910 platform(1p)
  │    ├─ run_distribute_train_for_ascend.sh   # launch distributed training with Ascend 910 platform(8p)
  │    ├─ run_standalone_train_for_gpu.sh      # launch standalone training with GPU platform(1p)
  │    ├─ run_eval_for_ascend.sh               # launch evaluating with Ascend 910 platform
  │    ├─ run_infer_310.sh                     # launch evaluating with Ascend 310 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 reading
  │    ├─ lr_generator.py                      # learning rate generator
  │    └─ llnet.py                             # network definition
  ├─ default_config.yaml                       # parameter configuration
  ├─ eval.py                                   # eval net
  ├─ export.py                                 # export checkpoint for ascend310_infer
  ├─ postprocess.py                            # post process for ascend310_infer
  ├─ preprocess.py                             # pre process for ascend310_infer
  ├─ requirements.txt                          # the pyaml package required by this network
  ├─ test.py                                   # the test for LLNet network
  ├─ train.py                                  # train net
  └─ write_mindrecords.py                      # write the mindrecords for train, eval, and test

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
'pretrain_epoch_size': 5,                      # the epoch number for pretrain
'finetrain_epoch_size': 300,                   # the epoch number for finetrain
'keep_checkpoint_max': 20,                     # max numbers to keep checkpoints
'save_ckpt_path': './',                        # save checkpoint path
'train_batch_size': 500,                       # input batch size for training
'val_batch_size': 1250,                        # input batch size for evaluation
'lr_init': [0.01, 0.01, 0.001, 0.001],         # initiate learning rate for the first three layer's pretrain and finetrain
'weight_decay': 0.0,                           # weight decay
'momentum': 0.9,                               # momentum

Training Process

Usage

# distribute training foar Ascend(8p)
bash run_distribute_train_ascend.sh [RANK_TABLE_FILE] [DATASET_PATH]
# standalone training for Ascend
bash run_standalone_train_for_ascend.sh [DEVICE_ID] [DATASET_PATH]

Launch

# distributed training example(8p) for Ascend
bash run_distribute_train_for_ascend.sh /home/hccl_8p_01234567.json /dataset
# standalone training example for for Ascend
bash 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] [DATASET_PATH] [CHECKPOINT]

Launch

# Evaluation with checkpoint
bash run_eval_for_ascend.sh 5 ../dataset ./ckpt_5/llnet-rank5-286_408.ckpt

Result

Evaluation result will be stored in the scripts path. Under this, you can find result like the followings in log.
PSNR=21.593(dB) SSIM=0.617

Inference Process

Export MindIR

Export MindIR on local

python export.py --device_target [PLATFORM] --device_id [DEVICE_ID] --checkpoint [CHECKPOINT_FILE] --file_format [FILE_FORMAT] --file_name [FILE_NAME]

The checkpoint_file parameter is required,
PLATFORM should be in ["Ascend", "GPU", "CPU"]
DEVICE_ID should be in [0-7]
FILE_FORMAT should be in ["AIR", "ONNX", "MINDIR"], the default value is MINDIR
FILE_NAME the base name for the exported model, the default value is llnet

Infer on Ascend310

Before performing inference, the mindir file must bu exported by export.py script. We only provide an example of inference using MINDIR model.
Current batch_Size can only be set to 1.

# Ascend310 inference
bash run_infer_310.sh [MINDIR_PATH] [DATASET_PATH] [NEED_PREPROCESS] [DEVICE_ID]

• MINDIR_PATH should be the filename of the MINDIR model.
• DATASET_PATH should be the path of the dataset.
• NEED_PREPROCESS can be y or n. It should by y for the first time of running.
• 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.
PSNR: 21.582 (dB)
SSIM: 0.604

Model description

Performance

Training Performance

Parameters	Ascend 910
Model Version	LLNet
Resource	Ascend 910
uploaded Date	07/23/2022 (month/day/year)
MindSpore Version	1.5.1
Dataset	dbimagenes
Training Parameters	default_config.yaml
Optimizer	Adam
Loss Function	MSE
Loss	0.0105
Total time	0 h 17 m 21 s 2ps
Checkpoint for Fine tuning	21.5 M(.ckpt file)

Inference Performance

Parameters	Ascend 910
Model Version	LLNet
Resource	Ascend 910
uploaded Date	07/23/2022 (month/day/year)
MindSpore Version	1.5.1
Dataset	dbimagenes
batch_size	1250
outputs	289 pixels reconstructed
Accuracy	PSNR = 21.593 SSIM = 0.617

ModelZoo Homepage

Please check the official homepage.