fhipol/deeplearning

Note: if training with CUDA, add +cu116 to the pytorch packages in requirements.txt

This repository implements the DeepLearning Project explained in https://docs.google.com/document/d/1uMPmAMymwjkmgRkB2MRZ2zPo3pCkN9A27Jm5yHuy2wQ/edit?usp=sharing and deployed in https://huggingface.co/spaces/fhipol/deeplearning

The main files are:

• —detector_model.py implements the ModelExecutor class used to train the models
• —reporter.py is used to report the loss plots and save the data from the training on detector_model.py
• —detector.py implements the interface to filter images with the trained model
• —app.py defines the user interface with Gradio

The main directories are:

• —humans and logos are the training sets with two classes, 1 (presence of the feature) and 0 (no present)
• —history data is the directory where reporter.py saves the data and the loss plots
• —trained_models where the last model trained state is saved for later use

For training the model run detectormodel.py, but that will override the last saved state on trainedmodels! Most of model configuration can also be changed there