A UNET Architecture designed in PyTorch for image segmentation.
This is a project that allows users to implement UNET in PyTorch without having to husstle the design of the architecture. It allows users to simple give dataset path and run a run.py file.
To get started simple clone this repo and install the requirements specified in the requirements.txt file.
git clone https://github.com/bibekebib/Py_UNET
cd Py_UNET
pip3 install requirements.txtAfter that there is a config.json file where you simply changes all the requirements as per you need.
{
"DataLoader":{
"csv":"True",
"csv_path":"outputs/test.csv",
"csv_train_path":"",
"csv_test_path":"",
"Image_folder":"False",
"image_folder_path":"",
"Split":"True"
},
"Data":{
"NChannels":3,
"Height":256,
"Width":256
},
"Model":{
"Train_Batch_size":16,
"Test_Batch_size":8,
"Optimizer":"Adam",
"learning_rate":1e-3,
"epochs":5
},
"Misc":{
"Generate_log":"True",
"Generate_plot":"True",
"Generate_report":"True",
"report_path":"outputs/report/summary_log.md"
},
"Report":{
"EdenAI_API_Key":"",
"output_folder":"outputs/report/project_report.md"
}
}
Make yourself a new EdenAI_API_Key at https://www.edenai.co/.
DataLoader:
csv: This flag indicates whether the data will be loaded from a CSV file. If set to "True," the data loader will use CSV files.
csv_path: The path to the main CSV file containing the data.
csv_train_path: Path to the CSV file for training data (if applicable).
csv_test_path: Path to the CSV file for testing data (if applicable).
Image_folder: This flag indicates whether the data will be loaded from image files. If set to "True," the data loader will use image files.
image_folder_path: The path to the folder containing the image data.
Split: This flag indicates whether the data will be split into training and testing sets. If set to "True," data splitting will occur.
Data:
NChannels: Number of color channels in the images (e.g., 3 for RGB).
Height: Height of the images in pixels.
Width: Width of the images in pixels.
Model:
Train_Batch_size: Batch size for training the model.
Test_Batch_size: Batch size for testing the model.
Optimizer: The optimizer algorithm to be used for model training (e.g., "Adam").
learning_rate: The learning rate used by the optimizer during training.
epochs: Number of epochs for which the model will be trained.
Misc:
Generate_log: This flag indicates whether a log will be generated during the process. If set to "True," a log will be created.
Generate_plot: This flag indicates whether plots will be generated (e.g., learning curves). If set to "True," plots will be created.
Generate_report: This flag indicates whether a report will be generated. If set to "True," a report will be created.
report_path: The path where the generated summary log will be saved.
Report:
EdenAI_API_Key: An API key for accessing EdenAI services.
output_folder: The folder where the project report will be saved in Markdown format.
After configuring config.json file, you can simple run run.py file and if your given path is legimitate, then it will start training.
python3 run.py
This generates a boiler plate for the research paper for you.
You can use it as your own boilerplate or just copy paste and modify according to your needs.
Just head over to outputs/reports/project_report.md and see a generated report paper just crafted by you.
This supports all the devices, CPU or CUDA or MPS. It will automatically detect and do as per the host device configuration.
It also generates a model summary log file to you. Check it out in outputs/reports/summary_log.md
If you Generate_plot=True, you get a loss graph in the outputs/graphs/ folder.
The final model will be saved in the outputs/model folder as model.pth file.
You can do prediction over you custom data simple heading to predict.py file and giving path to your file.
If you liked this tool, you are always welcome to make you contribution. Simply fork it and raise your issues.