Open pyni opened 6 years ago
Hi, Yes. Currently it is Yolo1 with darknet19. Direct location prediction helps anchor boxes work better. I do not intent to implement anchor boxes because I feel they are hard coded with prior scales and shapes of bounding boxes in the dataset. However, I do intend to combine image classification knowledge into detection as the paper proposed.
best, Wenxi
Hi, thanks for your code. But some important points for yolov2 are hard to be found in your code, such as anchor boxes, direct location prediction and so on. So i wonder whether your code is to combine yolo1 with darknet19 ?