Predict 报错ValueError: Cannot assign non-leaf Tensor to parameter 'weight'. Model parameters must be created explicitly. To express 'weight' as a function of another Tensor, compute the value in the forward() method.

[KeciaHH]

File "/home/hhee/Desktop/yolov8-pytorch/predict.py", line 91, in
r_image = yolo.detect_image(image, crop = crop, count=count)
File "/home/hhee/Desktop/yolov8-pytorch/yolo.py", line 144, in detect_image
outputs = self.net(images)
File "/home/hhee/anaconda3/envs/kws/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/home/hhee/anaconda3/envs/kws/lib/python3.9/site-packages/torch/nn/parallel/data_parallel.py", line 170, in forward
replicas = self.replicate(self.module, self.device_ids[:len(inputs)])
File "/home/hhee/anaconda3/envs/kws/lib/python3.9/site-packages/torch/nn/parallel/data_parallel.py", line 175, in replicate
return replicate(module, device_ids, not torch.is_grad_enabled())
File "/home/hhee/anaconda3/envs/kws/lib/python3.9/site-packages/torch/nn/parallel/replicate.py", line 148, in replicate
setattr(replica, key, param)
File "/home/hhee/anaconda3/envs/kws/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1632, in setattr
self.register_parameter(name, value)
File "/home/hhee/anaconda3/envs/kws/lib/python3.9/site-packages/torch/nn/modules/module.py", line 573, in register_parameter
raise ValueError(
ValueError: Cannot assign non-leaf Tensor to parameter 'weight'. Model parameters must be created explicitly. To express 'weight' as a function of another Tensor, compute the value in the forward() method.
加载权重成功了，infer报错

[ailongxia]

将nets\yolo.py下fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape))改成fusedconv.weight.data = torch.mm(w_bn, w_conv).view(fusedconv.weight.shape)；fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)改成fusedconv.bias.data = torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn即可

[Weichy9218]

Thanks a lot. I used ailongxia's solution to solve the same issue. Could you tell me the reason why it can deal with it?