How to configure channel size

[thhart]

I checked the training tutorial but could not find a central configuration setting to specify the input channel size for training. Is this possible within yaml files or is it necessary to change something in yolo.py for instance?
I tried to put ch into yaml but it looks like not being parsed.

[glenn-jocher]

@thhart you can create a YOLOv5 model with non-default channel size with PyTorch Hub. See PyTorch Hub tutorial:
https://docs.ultralytics.com/yolov5

The training dataloader defaults to 3 ch images, you'd have to manually modify it to your needs:

yolov5/utils/datasets.py

Line 336 in ab0db8d

class LoadImagesAndLabels(Dataset): # for training/testing

[glenn-jocher]

@thhart by the way, yaml['ch'] fields are not currently used, but you could modify yolo.py directly to use these similar to now yaml['nc'] is used or overridden.

yolov5/models/yolo.py

Lines 69 to 87 in ab0db8d

	class Model(nn.Module):
	def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None): # model, input channels, number of classes
	super(Model, self).__init__()
	if isinstance(cfg, dict):
	self.yaml = cfg # model dict
	else: # is *.yaml
	import yaml # for torch hub
	self.yaml_file = Path(cfg).name
	with open(cfg) as f:
	self.yaml = yaml.load(f, Loader=yaml.FullLoader) # model dict
	# Define model
	if nc and nc != self.yaml['nc']:
	logger.info('Overriding model.yaml nc=%g with nc=%g' % (self.yaml['nc'], nc))
	self.yaml['nc'] = nc # override yaml value
	self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch]) # model, savelist, ch_out
	self.names = [str(i) for i in range(self.yaml['nc'])] # default names
	# print([x.shape for x in self.forward(torch.zeros(1, ch, 64, 64))])

[glenn-jocher]

@thhart I would perhaps update yolo.py L84 to this to allow you to use a ch field in the yaml.

        self.model, self.save = parse_model(deepcopy(self.yaml), ch=[self.yaml.get('ch', ch)])  # model, savelist

[glenn-jocher]

@thhart PR #1741 is merged now, adding support for optional input channel definition in model yaml files, i.e.

# parameters
nc: 80  # number of classes
ch: 10  # input channels  <------------------------
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple

Tested with 1, 3 and 10 channel models.

[wangfurong123]

Hello, I used the YOLOv5 model to train a single-channel image, but it was unsuccessful. How can I modify the data loader part in the datasets.py file to train a single-channel image? Sincerely hope you can answer, thank you.

[glenn-jocher]

@wangfurong123 dataloaders are in datasets.py:

yolov5/utils/datasets.py

Lines 377 to 378 in 7a39803

	class LoadImagesAndLabels(Dataset):
	# YOLOv5 train_loader/val_loader, loads images and labels for training and validation

[xiaoche-24]

I have successfully obtained a six-channel input pt model. How can I convert it to an onnx model? What modifications need to be made to export.py?I ran export.py directly and reported the following error:
Traceback (most recent call last):
File "export.py", line 653, in
main(opt)
File "export.py", line 648, in main
run(**vars(opt))
File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "export.py", line 549, in run
y = model(im) # dry runs
File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 245, in forward
return self._forward_once(x, profile, visualize) # single-scale inference, train
File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 121, in _forward_once
x = m(x) # run
File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/lixiaojun/workfile/yolov5-master/models/common.py", line 3013, in forward_fuse
return self.act(self.conv(x))
File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 463, in forward
return self._conv_forward(input, self.weight, self.bias)
File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward
return F.conv2d(input, weight, bias, self.stride,
RuntimeError: Given groups=1, weight of size [32, 3, 6, 6], expected input[1, 6, 640, 640] to have 3 channels, but got 6 channels instead

[glenn-jocher]

Hi @xiaoche-24! To convert a six-channel input PyTorch model to an ONNX model, you need to make some modifications in the export.py script.

You received the error because the weights of one of the layers in your model have a fixed size of [32, 3, 6, 6]. This implies that it expects an input with 3 channels, but your input has 6 channels.

To resolve this issue, you need to modify the export.py script. Specifically, you should update the line where the model is defined to handle the case of your six-channel input. You can modify the parse_model function in models/yolo.py to accept a custom number of input channels.

Make sure to change the default value of ch to 6 (or the number of channels in your input) and re-run the export script.

Let me know if you have any further questions or need additional assistance!

[wq247726404]

I have successfully obtained a six-channel input pt model. How can I convert it to an onnx model? What modifications need to be made to export.py?I ran export.py directly and reported the following error: Traceback (most recent call last): File "export.py", line 653, in main(opt) File "export.py", line 648, in main run(**vars(opt)) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context return func(*args, **kwargs) File "export.py", line 549, in run y = model(im) # dry runs File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 245, in forward return self._forward_once(x, profile, visualize) # single-scale inference, train File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 121, in _forward_once x = m(x) # run File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/workfile/yolov5-master/models/common.py", line 3013, in forward_fuse return self.act(self.conv(x)) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 463, in forward return self._conv_forward(input, self.weight, self.bias) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: Given groups=1, weight of size [32, 3, 6, 6], expected input[1, 6, 640, 640] to have 3 channels, but got 6 channels instead

Hi bro, can I tweet you to ask some questions?

[xiaoche-24]

I have successfully obtained a six-channel input pt model. How can I convert it to an onnx model? What modifications need to be made to export.py?I ran export.py directly and reported the following error: Traceback (most recent call last): File "export.py", line 653, in main(opt) File "export.py", line 648, in main run(**vars(opt)) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context return func(*args, **kwargs) File "export.py", line 549, in run y = model(im) # dry runs File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 245, in forward return self._forward_once(x, profile, visualize) # single-scale inference, train File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 121, in _forward_once x = m(x) # run File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/workfile/yolov5-master/models/common.py", line 3013, in forward_fuse return self.act(self.conv(x)) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 463, in forward return self._conv_forward(input, self.weight, self.bias) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: Given groups=1, weight of size [32, 3, 6, 6], expected input[1, 6, 640, 640] to have 3 channels, but got 6 channels instead

Hi bro, can I tweet you to ask some questions?

sry，I don't have Twitter.You can ask the question directly here.

[wq247726404]

I have successfully obtained a six-channel input pt model. How can I convert it to an onnx model? What modifications need to be made to export.py?I ran export.py directly and reported the following error: Traceback (most recent call last): File "export.py", line 653, in main(opt) File "export.py", line 648, in main run(**vars(opt)) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context return func(*args, **kwargs) File "export.py", line 549, in run y = model(im) # dry runs File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 245, in forward return self._forward_once(x, profile, visualize) # single-scale inference, train File "/home/lixiaojun/workfile/yolov5-master/models/yolo.py", line 121, in _forward_once x = m(x) # run File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/workfile/yolov5-master/models/common.py", line 3013, in forward_fuse return self.act(self.conv(x)) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl return forward_call(*input, **kwargs) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 463, in forward return self._conv_forward(input, self.weight, self.bias) File "/home/lixiaojun/anaconda3/envs/yolov5/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 459, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: Given groups=1, weight of size [32, 3, 6, 6], expected input[1, 6, 640, 640] to have 3 channels, but got 6 channels instead

Hi bro, can I tweet you to ask some questions?

sry，I don't have Twitter.You can ask the question directly here.

do you have wechat xiongdi

[glenn-jocher]

@wq247726404 i'm here to help with any questions you may have regarding YOLOv5! Feel free to ask here and I'll do my best to assist you.