whria78 / data-in-paper-out Goto Github PK

Training CNN (Pytorch) and Deployment on the WWW (ONNX)

• Windows (or Linux)
• CPU processor (or GPU, CUDA)

• MobileNet v2
• EfficientNet Lite

• Download the code (ZIP)

• Get Anaconda version 3.8 or newer (https://www.anaconda.com/products/distribution)

  

  

  Please be sure to add the system PATH.

  

• Check the pretrained DEMO

  

  Run "0. requirement.bat" to install the required libraries.

  Run "2. run_demo_server.bat", and then "3. connect_demo_server.bat", to check the default DEMO.

• Train a custom model and try the DEMO

  Run "1. train_mobilenet.bat" if you want to train a CNN (MobileNet).

  Run "2. run_demo_server.bat", and then "3. connect_demo_server.bat", to check the customized DEMO.

The default DEMO shows the result of ensemble of efficientnet (/demo/model_eff_30e_0.onnx) and mobilenet (/demo/model_mob_30e_0.onnx). If you want to change or add more models, please check (/demo/dxinfo.js)

[LINUX] python train.py --model efficientnet --epoch 30 --step 10 --lr 0.005

[LINUX] python demo.py

[LINUX] xdg-open http://127.0.0.1:8000

The DEMO can be hosted on the WWW. Here is a web DEMO hosted by the github.

https://whria78.github.io/demo_onychomycosis/#

(A) Onychomycosis dataset came from the following paper.

Deep neural networks show an equivalent and often superior performance to dermatologists in onychomycosis diagnosis: Automatic construction of onychomycosis datasets by region-based convolutional deep neural network, PLOS One 2018

Paper : https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0191493

Dataset : https://figshare.com/articles/dataset/Model_Onychomycosis_Training_Datasets_JPG_thumbnails_and_Validation_Datasets_JPG_images_/5398573?file=9302506

(B) Clinical photographs of melanoma and nevus are available in the following link (CAN5600 dataset).

https://github.com/whria78/can