Implement a minimal YOLOv2 in Pytorch for learning purpose. The project focus on implementations of the logic code behind the YOLOv2 algorithms. Therefore, I would like to skip optional features as much as posible such as no loading config, uses only Darknet19 as a backbone network.
- Implement Darknet19
- Implement YOLO CNN using Darknet
- Implement Reog layer
- Implement region loss
- load pre-trained darknet19 weights
- load passcal voc 2012 dataset (generate input, label)
- Implement train YOLO
- Save and load trained model
- Implement inference
- Show loss graph using TensorBoard
- Implement jupyter notebook for training on Google Colab/kaggle