Converting YoloV5s to Pb file failed

[kaldarabsah]

I am getting this error message when I convert yolov5s.pt to pb file

Traceback (most recent call last):
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 1607, in _create_c_op
c_op = c_api.TF_FinishOperation(op_desc)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Depth of input (16) is not a multiple of input depth of filter (512) for 'conv_route4/Conv2D' (op: 'Conv2D') with input shapes: [?,?,?,16], [1,1,512,256].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/Users/kaldarabsah/Downloads/Yolov5_tf/core/yolov5.py", line 170, in init
self.conv_lbbox, self.conv_mbbox, self.conv_sbbox = self.__build_network(input_data)
File "/Users/kaldarabsah/Downloads/Yolov5_tf/core/yolov5.py", line 199, in __build_network
y19, layer_num = cspstage(input_data, self.trainable, 16init_width_size, init_depth_size, layer_num, 4, 1+7init_depth_size)
File "/Users/kaldarabsah/Downloads/Yolov5_tf/core/yolov5.py", line 92, in cspstage
route = conv(route, (1, 1, c, c / 2), trainable=trainable, name='conv_route%d' % route_nums, act_fun='mish')
File "/Users/kaldarabsah/Downloads/Yolov5_tf/core/yolov5.py", line 33, in conv
conv = tf.nn.conv2d(input=input_data, filter=weight, strides=strides, padding=padding)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/ops/nn_ops.py", line 2010, in conv2d
name=name)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/ops/gen_nn_ops.py", line 1071, in conv2d
data_format=data_format, dilations=dilations, name=name)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/framework/op_def_library.py", line 794, in _apply_op_helper
op_def=op_def)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/util/deprecation.py", line 507, in new_func
return func(*args, **kwargs)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 3357, in create_op
attrs, op_def, compute_device)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 3426, in _create_op_internal
op_def=op_def)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 1770, in init
control_input_ops)
File "/opt/miniconda2/envs/airflow/lib/python3.7/site-packages/tensorflow_core/python/framework/ops.py", line 1610, in _create_c_op
raise ValueError(str(e))
ValueError: Depth of input (16) is not a multiple of input depth of filter (512) for 'conv_route4/Conv2D' (op: 'Conv2D') with input shapes: [?,?,?,16], [1,1,512,256].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "from_darknet_weights_to_pb.py", line 18, in
model = YOLOV5(input_data, trainable=False)
File "/Users/kaldarabsah/Downloads/Yolov5_tf/core/yolov5.py", line 172, in init
raise NotImplementedError('Can not build up yolov5 network')
NotImplementedError: Can not build up yolov5 network

Kindly, add an example how to convert pt file to pb

Thank you.

[avBuffer]

OK, Tks! I try it.

[kaldarabsah]

Did it work for YoloV5s.pt file?

[NanoCode012]

@kaldarabsah @avBuffer , hello, as I see it, the convert script from_darknet_weights_to_pb.py only works from .weights to .pb right? How can we use a .pt file?

From #2 (comment), you recommended to convert to onnx to tf. Can you recommend me a repo? We have to convert from onnx to tf1 right, not tf2?

[kaldarabsah]

@NanoCode012 . I tried to add YoloV5 to the script but when I ran it, it did not work.
https://github.com/avBuffer/Yolov5_tf/blob/master/scripts/from_darknet_weights_to_pb.py#L3

YoloV5 only has pt file. Here are the weights for Yolov5 from the original repo https://drive.google.com/drive/folders/1Drs_Aiu7xx6S-ix95f9kNsA6ueKRpN2J

[NanoCode012]

@NanoCode012 . I tried to add YoloV5 to the script but when I ran it, it did not work.
https://github.com/avBuffer/Yolov5_tf/blob/master/scripts/from_darknet_weights_to_pb.py#L3

@kaldarabsah Yes, because when I read it, it does not convert yolov5. Only darknet .weights.

Yolov5_tf/scripts/from_darknet_weights_to_pb.py

Line 8 in 682db0d

darknet_weights = 'model_data/yolov4.weights'

[avBuffer]

Hey kaldarabsah, Thanks! I fixed the error, you can try it!

[lengyuner]

@kaldarabsah
hi, where did you get the yolov5.weight, the link is invalid.

TFT