DetectX is a project dedicated to re-implementing and training multiple foundational object detection architectures from scratch. The goal is to understand the inner workings and unique strengths of various detection models by building them from the ground up.
The primary objectives of DetectX are:
- To re-implement several well-known object detection architectures from scratch on Pascal VOC.
- To train and evaluate each model on relevant datasets, providing a hands-on comparison of their performance and capabilities.
DetectX will include implementations of the following architectures:
- YOLO (You Only Look Once)
- DETR (Detection Transformer)
- RetinaNet
- Faster R-CNN
- CornerNet