mindspore/model_zoo/official/cv/FCN8s

• Contents
• FCN Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
• • Script and Sample Code
  • Script Parameters
  • Data Generation
  • • Training Data
  • Training Process
  • • Training
  • Evaluation Process
  • • Evaluation
• Model Description
• • Performance
  • • Evaluation Performance
    • Inference Performance
    • • FCN8s on PASCAL VOC
• Description of Random Situation
• ModelZoo Homepage

Contents

• FCN Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
  • Script and Sample Code
  • Script Parameters
  • Data Generation
    • Training Data
  • Training Process
    • Training
  • Evaluation Process
    • Evaluation
• Model Description
  • Performance
    • Evaluation Performance
    • Inference Performance
• Description of Random Situation
• ModelZoo Homepage

FCN Description

FCN is mainly used in the field of image segmentation, which is an end-to-end segmentation method. FCN changes the last full connected layers of VGG to process images of any size, reduce the parameters and improve the segmentation speed of the model. FCN uses VGG structure in the encoding part and deconvolution / up sampling operation in the decoding part to recover the image resolution. Finally, FCN8s uses 8 times deconvolution / up sampling operation to restore the output image to the same size as the input image.

[Paper]: Long, Jonathan, Evan Shelhamer, and Trevor Darrell. "Fully convolutional networks for semantic segmentation." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015.

Model Architecture

FCN8s uses VGG16 without the full connected layers as the encoding part, and fuses the features of the 3rd, 4th and 5th pooling layers in VGG16 respectively. Finally, the deconvolution of stride 8 is used to obtain the segmented image.

Dataset

Dataset used:

PASCAL VOC 2012

Environment Requirements

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

Quick Start

After installing MindSpore through the official website, you can start training and evaluation by following these steps：

• running on Ascend with default parameters

  # run training example
  python train.py --device_id device_id
  
  # run evaluation example with default parameters
  python eval.py --device_id device_id
  

Script Description

Script and Sample Code

├── cv
    ├── FCN8s
        ├── README.md                 // descriptions about FCN
        ├── scripts
            ├── run_train.sh
            ├── run_standalone_train.sh
            ├── run_eval.sh
            ├── build_data.sh
        ├── src
        │   ├──data
        │       ├──build_seg_data.py       // creating dataset
        │       ├──dataset.py          // loading dataset
        │   ├──nets
        │       ├──FCN8s.py            // FCN-8s architecture
        │   ├──loss
        │       ├──loss.py            // loss function
        │   ├──utils
        │       ├──lr_scheduler.py            // getting learning_rateFCN-8s  
        ├── train.py                 // training script
        ├── eval.py                  //  evaluation script

Script Parameters

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

• config for FCN8s

     # dataset
    'data_file': '/data/workspace/mindspore_dataset/FCN/FCN/dataset/MINDRECORED_NAME.mindrecord', # path and name of one mindrecord file
    'batch_size': 32,
    'crop_size': 512,
    'image_mean': [103.53, 116.28, 123.675],
    'image_std': [57.375, 57.120, 58.395],
    'min_scale': 0.5,
    'max_scale': 2.0,
    'ignore_label': 255,
    'num_classes': 21,
  
    # optimizer
    'train_epochs': 500,
    'base_lr': 0.015,
    'loss_scale': 1024.0,
  
    # model
    'model': 'FCN8s',
    'ckpt_vgg16': '',
    'ckpt_pre_trained': '',
  
    # train
    'save_steps': 330,
    'keep_checkpoint_max': 5,
    'ckpt_dir': './ckpt',
  

For more information, see config.py.

Data Generation

Training Data

• build mindrecord training data

  sh build_data.sh
  or
  python src/data/build_seg_data.py  --data_root=/home/sun/data/Mindspore/benchmark_RELEASE/dataset  \
                                     --data_lst=/home/sun/data/Mindspore/benchmark_RELEASE/dataset/trainaug.txt  \
                                     --dst_path=dataset/MINDRECORED_NAME.mindrecord  \
                                     --num_shards=1  \
                                     --shuffle=True
  

Training Process

Training

• running on Ascend with default parameters

  python train.py --device_id device_id
  

  Checkpoints will be stored in the default path

Evaluation Process

Evaluation

• Evaluated on Pascal VOC 2012 validation set using Ascend

  Before running the command, check the path of the checkpoint used for evaluation. Please set the absolute path of the checkpoint

  python eval.py
  

  After running the above command, you can see the evaluation results on the terminal. The accuracy on the test set is presented as follows:

  mean IoU  0.6425
  

Model Description

Performance

Evaluation Performance

Parameters	Ascend
Model Version	FCN-8s
Resource	Ascend 910; CPU 2.60GHz, 192cores; Memory 755G; OS Euler2.8
uploaded Date	12/30/2020 (month/day/year)
MindSpore Version	1.1.0-alpha
Dataset	PASCAL VOC 2012
Training Parameters	epoch=500, steps=330, batch_size = 32, lr=0.015
Optimizer	Momentum
Loss Function	Softmax Cross Entropy
outputs	probability
Loss	0.038
Speed	1pc: 564.652 ms/step;
Scripts	FCN script

Inference Performance

FCN8s on PASCAL VOC

Parameters	Ascend
Model Version	FCN-8s
Resource	Ascend 910; OS Euler2.8
Uploaded Date	12/30/2020 (month/day/year)
MindSpore Version	1.1.0-alpha
Dataset	PASCAL VOC 2012
batch_size	16
outputs	probability
mean IoU	64.25

Description of Random Situation

We set the random seeds in train.py.

ModelZoo Homepage

Please check the official homepage.