error when testing forward pass about yolo_v3_tutorial_from_scratch

[allenwu5]

HI @Anikily ,
I can reproduce the error you encountered on my PC (os: windows10, cuda: available).
Then I do the debug, where x_y_offset.is_cuda = True, but prediction.is_cuda = False.

So I think you might be right, one in GPU but the other in CPU.
Then I added few lines in darknet.py (just pick somewhere looks suitable ...) so that prediction.is_cuda can be True later.

                # Transform
                x = x.data
                if CUDA:
                    x = x.cuda()

Now it pass with exit code 0.

But I still not sure where is a good place to put x = x.cuda() ...

[Anikily]

Hi @arashilen , thansks for your reply.I have solved this problem yesterday in a similar way to yours.And I did this step in util.py...in just ablove the line which reported the error.

[allenwu5]

Hi @Anikily ,
Maybe you can debug value of x["anchors"] first.
E.g. my case would be this:
x["anchors"] = '10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326'

which is come from https://github.com/ayooshkathuria/YOLO_v3_tutorial_from_scratch/blob/master/cfg/yolov3.cfg

os: macos

[Anikily]

thanks ,I have solved this problem, and I encountered another problem :
Traceback (most recent call last):
File "darknet.py", line 238, in
pred = model(inp,torch.cuda.is_available())
File "/home/tensor/anaconda2/envs/py36/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "darknet.py", line 227, in forward
x = predict_transform(x,inp_dim,anchors,num_classes,CUDA)
File "/home/tensor/jupyter/luyu/yolo3/util.py", line 79, in predict_transform
prediction[:,:,:2] += x_y_offset
RuntimeError: Expected object of type torch.FloatTensor but found type torch.cuda.FloatTensor for argument #4 'other'
maybe it means that the type of x_y_offset is torch,cuda.FloatTensor,but the type of prediction is torch.FloatTensor and they are not at the same memory address.How should I do?Transform the type of prediction to Cuda?