RuntimeError: The expanded size of the tensor (3) must match the existing size (864) at non-singleton dimension 3. Target sizes: [32, 3, 3, 3]. Tensor sizes: [864]

[DL-Alva]

Sorry for troubling you. When I run train_meta.py and load weightfile, a runtimeerror occured:

logging to backup/metayolo_novel0_neg1 class_scale 1

RuntimeErrorTraceback (most recent call last)

in () 14 region_loss = model.loss 15 ---> 16 model.load_weights(weightfile) 17 model.print_network() ~/lkj项目/FSD_yolo/darknet_meta.py in load_weights(self, weightfile) 376 batch_normalize = int(block['batch_normalize']) 377 if batch_normalize: --> 378 start = load_conv_bn(buf, start, model[0], model[1]) 379 else: 380 ~/lkj项目/FSD_yolo/cfg.py in load_conv_bn(buf, start, conv_model, bn_model) 453 bn_model.running_mean.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b 454 bn_model.running_var.copy_(torch.from_numpy(buf[start:start+num_b])); start = start + num_b --> 455 conv_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_w])); start = start + num_w 456 return start 457 RuntimeError: The expanded size of the tensor (3) must match the existing size (864) at non-singleton dimension 3. Target sizes: [32, 3, 3, 3]. Tensor sizes: [864] Do you know what's wrong with this? Thank you so much.

[DL-Alva]

The folder "backup" is created manually. Is that right?

[bingykang]

Yes.

[DL-Alva]

Hi, thank you for your quickly reply. But I haven't solved the error mentioned above till now. Is it related with Broadcasting semantics in Pytorch? Thanks a lot.

[bingykang]

Are you using the cfg files in the instruction?

[DL-Alva]

Yes, I use cfg files and change the path of train set and val set.

[PCH10507323]

Maybe you can refer to this. https://github.com/marvis/pytorch-yolo3/issues/15

[DL-Alva]

Thank you for your opinions. I have tried these methods mentioned in the link above but solutions didn't work.My torch version is 1.3.0.

[bingykang]

Our code is implemented with PyTorch 0.3.1 and Python 2.7. You may want to try this version.

[DL-Alva]

Thank you for your suggestion! I think I should try it.

[christegho]

I had the same issue, so I switched to PyTorch 0.3.1 but now I get

    import torch.utils.checkpoint as cp
ImportError: No module named checkpoint

@DL-Alva Were you able to resolve the error with the The expanded size of the tensor (3) ?

if yes, how and what version of Torch did you end up using?

[christegho]

This ended up working, although not sure what I did to resolve the issue.

I did install torchvision==0.4.2, and installed torch 0.4.0 then reinstalled torch 0.3.1.

[shenglih]

torch==0.3.1, python2.7, torchvision==0.2.0 work

[90000mileswind]

Sorry for troubling you. When I run train_meta.py and load weightfile, a runtimeerror occured:

logging to backup/metayolo_novel0_neg1

class_scale 1 RuntimeErrorTraceback (most recent call last) in () 14 region_loss = model.loss 15 ---> 16 model.load_weights(weightfile) 17 model.print_network()

~/lkj项目/FSD_yolo/darknet_meta.py in load_weights(self, weightfile) 376 batch_normalize = int(block['batch_normalize']) 377 if batch_normalize: --> 378 start = load_conv_bn(buf, start, model[0], model[1]) 379 else: 380

~/lkj项目/FSD_yolo/cfg.py in load_conv_bn(buf, start, conv_model, bn_model) 453 bn_model.runningmean.copy(torch.from_numpy(buf[start:start+num_b])); start = start + num_b 454 bn_model.runningvar.copy(torch.from_numpy(buf[start:start+num_b])); start = start + num_b --> 455 convmodel.weight.data.copy(torch.from_numpy(buf[start:start+num_w])); start = start + num_w 456 return start 457

RuntimeError: The expanded size of the tensor (3) must match the existing size (864) at non-singleton dimension 3. Target sizes: [32, 3, 3, 3]. Tensor sizes: [864]

Do you know what's wrong with this? Thank you so much.

这个问题我觉得可以修改，cfg.py第456行.copy_()如下： size_conv = conv_model.weight.data.size() # 这两行改过，将被复制的shape改成一致的 convmodel.weight.data.copy(torch.from_numpy(buf[start:start+num_w]).view(size_conv)); start = start + num_w

copy_()复制和被复制的shape不一样，所有如果改了应该就可以。但是，我还遇到其他一些问题，或许换了pytorch的版本能解决问题！

[Fly-dream12]

Have you solved this problem? @DL-Alva @90000mileswind I also met this problem. Thanks for your help.

[MichaelXin]

Same errors. I use cuda8, pytorch 0.3.1, python 2.7 and torchvision 0.2.0. Have you solved this problem? @DL-Alva

[JakeZDK]

@90000mileswind 你好，我也是同样的问题。view(size_conv)是否与存储的预训练数据顺序一致呢？

[NHW2017]

Our code is implemented with PyTorch 0.3.1 and Python 2.7. You may want to try this version.

Hello, can you tell me the CUDA version you used in the test?

[mountain111]

It is OK using pytorch0.3.1-py2.7

[VinhLoiIT]

Open cfg.py file at line 455 and change to

conv_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_w]).reshape_as(conv_model.weight.data)); start = start + num_w 

worked for me

[binbinlan]

Open cfg.py file at line 455 and change to

conv_model.weight.data.copy_(torch.from_numpy(buf[start:start+num_w]).reshape_as(conv_model.weight.data)); start = start + num_w 

worked for me that really works!! thanks!!!!!!