yolov4 export onnx get wrong

[zys1994]

when i train yolov4 ,i replace mish with leaky. when i convert the model to onnx,it get wrong like that.
my environment is pytorch1.5,onnx-1.7

Namespace(agnostic_nms=False, augment=False, cfg='pre_model/yolov4-facemask-leakly.cfg', classes=None, conf_thres=0.3, device='', fourcc='mp4v', half=False, img_size=416, iou_thres=0.6, names='pre_model/obj.name', output='output', save_txt=False, source='data/samples', view_img=False, weights='pre_model/yolov4-facemask-leakly_best.weights')
Using CPU

Fusing layers...
/home/bill.zeng/dl_base/object-det/yolov3-pytorch/utils/layers.py:60: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if nx == na:  # same shape
Traceback (most recent call last):
  File "detect.py", line 187, in <module>
    detect()
  File "detect.py", line 48, in detect
    input_names=['images'], output_names=['classes', 'boxes'])
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/onnx/__init__.py", line 168, in export
    custom_opsets, enable_onnx_checker, use_external_data_format)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/onnx/utils.py", line 69, in export
    use_external_data_format=use_external_data_format)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/onnx/utils.py", line 488, in _export
    fixed_batch_size=fixed_batch_size)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/onnx/utils.py", line 334, in _model_to_graph
    graph, torch_out = _trace_and_get_graph_from_model(model, args, training)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/onnx/utils.py", line 291, in _trace_and_get_graph_from_model
    torch.jit._get_trace_graph(model, args, _force_outplace=False, _return_inputs_states=True)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 278, in _get_trace_graph
    outs = ONNXTracedModule(f, _force_outplace, return_inputs, _return_inputs_states)(*args, **kwargs)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 361, in forward
    self._force_outplace,
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/jit/__init__.py", line 348, in wrapper
    outs.append(self.inner(*trace_inputs))
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 548, in __call__
    result = self._slow_forward(*input, **kwargs)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 534, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "/home/bill.zeng/dl_base/object-det/yolov3-pytorch/models.py", line 238, in forward
    return self.forward_once(x)
  File "/home/bill.zeng/dl_base/object-det/yolov3-pytorch/models.py", line 292, in forward_once
    x = module(x)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 548, in __call__
    result = self._slow_forward(*input, **kwargs)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py", line 534, in _slow_forward
    result = self.forward(*input, **kwargs)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/modules/upsampling.py", line 131, in forward
    return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners)
  File "/home/bill.zeng/softmare/anaconda3/lib/python3.6/site-packages/torch/nn/functional.py", line 2991, in interpolate
    scale_factor_list[0], scale_factor_list[1])
RuntimeError: Input and output sizes should be greater than 0, but got input (H: 10, W: 6) output (H: 0, W: 0)

[nightsnack]

same issue, cfg: csresnext50-panet-spp

[zjd1988]

Hi @zys1994 @nightsnack ， I have solved this, you can replace the same files with attachment

models.zip

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove Stale label or comment or this will be closed in 5 days.

[dengfenglai321]

Hi @zys1994 @nightsnack ， I have solved this, you can replace the same files with attachment
models.zip

hi thansk for you file.
I transfer to ONNX sucesseful.

and I have some question.
ONNX model only have two output:
box (10467, 4)
class(10467, 2) (I train two class).

how to decode this output to get final result?
in otherwords.
How I use nms to get final result?

could you share code?

thanks

[zjd1988]

@cendelian you can refer to this issue #1163

[huyhoangle86]

@zjd1988 @cendelian How do you guys convert models to onnx ? how to execute the program to get Onnx ? plz

[glenn-jocher]

Hello @huyhoangle86, @cendelian, and @nightsnack,

To convert YOLOv3 models to ONNX format, you typically need to use the export.py script provided in the repository. The process involves specifying the weights file and the configuration file for your model. Here's a general outline of the steps you would follow:

1. Ensure your environment is set up with the correct versions of PyTorch and ONNX that are compatible with each other.
2. Modify the export.py script if necessary to accommodate any custom changes you've made to the model architecture.
3. Run the export.py script with the appropriate arguments, such as --weights for your trained weights and --cfg for your model configuration.

For non-maximum suppression (NMS) and decoding the outputs, you would typically use the functions provided in the repository that handle these tasks. The outputs from the ONNX model will be raw predictions, which need to be processed by NMS to filter out overlapping bounding boxes and select the most probable detections.

If you encounter specific issues or errors during this process, please provide the error messages or describe the problem in detail, and I'll do my best to assist you.

For further guidance on exporting models and post-processing, you can refer to the Ultralytics documentation.

Keep in mind that sharing code directly through GitHub issues is not recommended due to potential licensing and compatibility issues. It's best to use the official repository code and documentation as a reference.