Code used by Felix during Pixmoving Hackathon. This repo just contains the code but not the dataset nor the trained models.

Here is a high level description of the important files:

d2/manage.py - Default manage.py file

d2/manage_enhanced.py - Include support for Stacked Frame, and Time Sequence Frames (for LSTM) imported as parts

d2/manage_enhanced_cv.py - Include Image Thresholding to eliminate light glare, also implemented as parts

d2/train/datasets.py - Dataloader to prepare training data for default model, stacked frame model, and LSTM model.

d2/train/train.py - Train various models such as LSTM, fine-tune Q model from RL, default CNN with grayscale stacked frame, etc

donkeycar/donkeycar/parts/cv.py - CV related algorithms implemented as parts. Image Thresholding, Stacked Frame, and Time Sequence Frames code can be found here

donkeycar/donkeycar/parts/keras.py - Wrap Keras models (such as LSTM, Q model, default CNN) as parts

Run on Jetson Nano following this guide.