Modify Deep Learning Scripts
Budget: $30 – $250 USD
1. Take sections of the full screen view (left and right corner) at runtime and present it to the model. You can take a single frame and split out the two corner views in training.You could also do this inside the eras model itself. Ultimately, we want to construct 3 different views at training run time. If neither of these can be done fast, you can pre-create different data directories with the sliced out pictures, but that is not ideal/
2. The model needs to be modified to take in 3 inputs which means there needs to be 3 convolutional streams to deal with the three datasets
3. You should concatenate images into one long rectangular picture and pass it into model.
To record data, use data_recorder.py. For pre-process images from data_recorder.py use generate_dataset_dirt5.py. To modify screen sizing use grab screen.py. To modify model, look at model_lib and model_trainner scripts and handler_tools
2. The model needs to be modified to take in 3 inputs which means there needs to be 3 convolutional streams to deal with the three datasets
3. You should concatenate images into one long rectangular picture and pass it into model.
To record data, use data_recorder.py. For pre-process images from data_recorder.py use generate_dataset_dirt5.py. To modify screen sizing use grab screen.py. To modify model, look at model_lib and model_trainner scripts and handler_tools