torch-classifier

This is a pytorch implementation of image classification in pytorch. You can use the models from the list belows:

• ResNet
• VGG
• Inception
• MobileNet
• ResNext

This repository includes:

• Training and Evaluation Codes
• Streamlit Classification Application

Streamlit Classification Application

Installation

1. Install docker. See Install Docker Engine on Ubuntu.

2. Install torch-classifier

git clone https://github.com/ryoherisson/torch-classifier.git

3. Build docker image

cd torch-classifier
sh ./docker/build.sh

4. Run docker container

sh ./docker/run.sh
cd /workspace

5. Setup

python setup.py develop --user

Usage

Create a configuration file based on configs/default.yml.

### Configs ###

# Data Configs
data:
  dataroot: ./data/sample/images/
  labelroot:
    train: ./data/sample/labels/train.csv
    val: ./data/sample/labels/test.csv
    test: ./data/sample/labels/test.csv
  img_size: [32, 32]
  n_channels: 3
  classes: ['airplane', 'bird', 'car', 'cat', 'deer', 'dog', 'horse', 'monkey', 'ship', 'truck']
  color_mean: [0.4914, 0.4822, 0.4465]
  color_std: [0.2023, 0.1994, 0.2010]

# Training Configs
train:
  batch_size: 128
  epochs: 30
  optimizer:
    type: sgd
    lr: 0.001
    momentum: 0.9
    decay: 0.0001
  criterion:
    type: cross_entropy
  n_gpus: 1
  save_ckpt_interval: 10

# Model Configs
model:
  name: resnet18
  n_classes: 10
  pretrained: False
  resume:
    # e.g) resume: ./logs/2020-07-26T00:19:34.918002/ckpt/best_acc_ckpt.pth if resume. Blank if not resume

# Other Configs
util:
  logdir: ./logs/

Prepare Dataset

Prepare a directory with the following structure:

datasets/
├── images
│   ├── airplane1.png
│   ├── car1.png
│   ├── cat1.png
│   └── deer1.png
└── labels
    ├── train.csv
    └── test.csv

The content of the csv file should have the following structure.

filename,label
airplane1.png,0
car1.png,1
cat1.png,3
deer1.png,4

An example of a dataset can be found in the dataset folder.

Train

python train.py --config ./configs/default.yml

Evaluation

python train.py --config ./configs/default.yml --eval

Tensorboard

Docker build

sh docker/tensorboard_build.sh

Run tensorboard

sh docker/tensorboard_run.sh

Access localhost:6006

Streamlit Application

streamlit run app/app.py -- --configfile {your-config-path}

You can now view your Streamlit app in your browser.
  Network URL: http://172.17.0.2:8501
  External URL: http://113.33.218.178:8501

Output of train and evaluation

You will see the following output in the log directory specified in the Config file.

logs/
└── 2020-07-26T14:21:39.251571
    ├── ckpt
    │   ├── best_acc_ckpt.pth
    │   ├── epoch0000_ckpt.pth
    │   └── epoch0001_ckpt.pth
    ├── metric
    │   ├── train_metric.csv
    │   ├── eval_metric.csv
    │   └── eval_cmx.png
    ├── tensorboard
    │   └── events.out.tfevents.1595773266.c47f841682de
    └── info.log

The contents of train_metric.csv and eval_metric.csv look like as follows:

epoch,train loss,train accuracy,train precision,train recall,train micro f1score
0,0.024899629971981047,42.832,0.4283200204372406,0.4215449392795563,0.42248743772506714
1,0.020001413972377778,54.61,0.5461000204086304,0.5404651761054993,0.5422631502151489

You will get loss, accuracy, precision, recall, micro f1score during training and as a result of evaluation.

The content of eval_cmx.png is a confusion matrix.