With some fine-tuning, my model finally arrived at a performance of 96.1% for AP@0.7. To execute my work, executing train.py and main.py should get you the result! If you would to skip the training process and just want to run inference, contact me and I'll send it over!