search

Zhongdao / Towards-Realtime-MOT

Joint Detection and Embedding for fast multi-object tracking
MIT License
2.37k stars 541 forks source link

Mismatch between model and weights #121

Closed denverdash closed 4 years ago

denverdash commented 4 years ago

I am getting the following error: builtins.RuntimeError: Error(s) in loading state_dict for Darknet: size mismatch for classifier.weight: copying a param with shape torch.Size([14455, 512]) from checkpoint, the shape in current model is torch.Size([30, 512]). size mismatch for classifier.bias: copying a param with shape torch.Size([14455]) from checkpoint, the shape in current model is torch.Size([30]).

It looks like the weights are not consistent with the model architecture. I noticed the weights are "jde." while the cfg is "yolov3" - is there a jde*.cfg I need to use?

The cfg I am using looks like: Namespace(cfg='cfg/yolov3_1088x608.cfg', conf_thres=0.5, input_video='../yolov3/videos/amazon-1088x608.mp4', iou_thres=0.5, min_box_area=200, nms_thres=0.4, output_format='video', output_root='./output/', track_buffer=30, weights='./models/jde.1088x608.uncertainty.pt')

The stack trace is: File "/home/denver/work/Towards-Realtime-MOT/demo.py", line 84, in track(opt) File "/home/denver/work/Towards-Realtime-MOT/demo.py", line 59, in track save_dir=frame_dir, show_image=False, frame_rate=frame_rate) File "/home/denver/work/Towards-Realtime-MOT/track.py", line 79, in eval_seq tracker = JDETracker(opt, frame_rate=frame_rate) File "/home/denver/work/Towards-Realtime-MOT/tracker/multitracker.py", line 164, in init self.model.load_state_dict(torch.load(opt.weights, map_location='cpu')['model'], strict=False) File "/home/denver/work/venvs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 830, in load_state_dict self.class.name, "\n\t".join(error_msgs)))

builtins.RuntimeError: Error(s) in loading state_dict for Darknet: size mismatch for classifier.weight: copying a param with shape torch.Size([14455, 512]) from checkpoint, the shape in current model is torch.Size([30, 512]). size mismatch for classifier.bias: copying a param with shape torch.Size([14455]) from checkpoint, the shape in current model is torch.Size([30]).

imchwan commented 4 years ago

I don't know the meaning of nID exactly, however, I modified code at multitracker.py. (https://github.com/Zhongdao/Towards-Realtime-MOT/blob/master/tracker/multitracker.py#161) self.model = Darknet(opt.cfg , nID=14455) or self.model = Darknet(opt.cfg) # default: nID = 0 size mismatch was solved.

denverdash commented 4 years ago

14455

Thanks that fixed it!

denverdash commented 4 years ago

@Zhongdao - I created a PR to fix this and a few other bugs I found: https://github.com/Zhongdao/Towards-Realtime-MOT/pull/122

mheriyanto commented 3 years ago

I don't know the meaning of nID exactly, however, I modified code at multitracker.py. (https://github.com/Zhongdao/Towards-Realtime-MOT/blob/master/tracker/multitracker.py#161) self.model = Darknet(opt.cfg , nID=14455) or self.model = Darknet(opt.cfg) # default: nID = 0 size mismatch was solved.

Thank you @imchwan . It works for me.