argusswift / YOLOv4-pytorch

This is a pytorch repository of YOLOv4, attentive YOLOv4 and mobilenet YOLOv4 with PASCAL VOC and COCO
1.68k stars 329 forks source link

YoloV4Loss __main__ does not work #213

Open a-haja opened 3 years ago

a-haja commented 3 years ago

@argusswift

The current implementation of the main function in YoloV4Loss is unfortunately not correct. You need to adjust the following: https://github.com/argusswift/YOLOv4-pytorch/blob/d0a3b6553c22a7e45e218fb1b82465367e833792/model/loss/yolo_loss.py#L163 instead of from model.build_model import Yolov4 and net = Yolov4() you should import Build_Model like this: from model.build_model import Build_Model; net = Build_Model()

###################################################################################################### I would also like to ask some questions in this regards: 1) What is mix in label_sbbox? (label_mix = label[..., 5:6]) https://github.com/argusswift/YOLOv4-pytorch/blob/d0a3b6553c22a7e45e218fb1b82465367e833792/model/loss/yolo_loss.py#L48 By the way: In https://github.com/argusswift/YOLOv4-pytorch/blob/d0a3b6553c22a7e45e218fb1b82465367e833792/utils/utils.py#L184 The GT label has only a size of 5 [class, x, y, w, h] and not [x+y+w+h+conf+mix]. We you add an extra value (mix) to the label array?

2) What is 150 in sbboxes? https://github.com/argusswift/YOLOv4-pytorch/blob/d0a3b6553c22a7e45e218fb1b82465367e833792/model/loss/yolo_loss.py#L52

Thank you in advance.