ValueError: Invalid output layout or shape with converted ONNX model when running validate_yolo.py

[Kylemaycry]

I am converting a .weight model, first to .h5 and then to .pb. Both are successful and can be verified by the validate_yolo.py
(In addition, can this repo convert .h5 to a TF savedmodel? )
The conversion to .onnx is also successful, but model cannot be verified with validate_yolo.py. The error message is as following:
Thanks in advance!!

The device support i8sdot:0, support fp16:0, support i8mm: 0
2023-11-27 21:24:16.453009: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2023-11-27 21:24:16.453046: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2023-11-27 21:24:19.901064: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/lib/python3.8/dist-packages/cv2/../../lib64:
2023-11-27 21:24:19.901113: W tensorflow/stream_executor/cuda/cuda_driver.cc:326] failed call to cuInit: UNKNOWN ERROR (303)
2023-11-27 21:24:19.901152: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (art-ipc427d): /proc/driver/nvidia/version does not exist
(416, 416)
2023-11-27 21:24:20.029914655 [W:onnxruntime:Default, upsamplebase.h:101 UpsampleBase] tf_half_pixel_for_nn is deprecated since opset 13, yet this opset 13 model uses the deprecated attribute
NCHW input layout
Traceback (most recent call last):
File "./tools/evaluation/validate_yolo.py", line 520, in
main()
File "./tools/evaluation/validate_yolo.py", line 511, in main
validate_yolo_model_onnx(model, image_file, anchors, class_names, args.elim_grid_sense, args.v5_decode, args.loop_count, args.output_path)
File "./tools/evaluation/validate_yolo.py", line 335, in validate_yolo_model_onnx
raise ValueError('invalid output layout or shape')
ValueError: invalid output layout or shape

The output of print(output_tensors[0],output_tensors[1]):
NodeArg(name='conv2d_9', type='tensor(float)', shape=['unk__308', 30, 0, 0]) NodeArg(name='conv2d_12', type='tensor(float)', shape=['unk__311', 30, 0, 0])

[david8862]

@Kylemaycry how did you convert to the .onnx model?

[Kylemaycry]

@Kylemaycry how did you convert to the .onnx model?
Hi @david8862, I used the following commands, found them here:https://github.com/david8862/keras-YOLOv3-model-set#onnx-model-convert
Thanks the help!

export TF_KERAS=1
python3 tools/model_converter/keras_to_onnx.py
--keras_model_file=${MODEL_FOLDER}/yolov3-tiny.h5
--output_file=${MODEL_FOLDER}/yolo-v3-tiny.onnx
--op_set=15 --inputs_as_nchw

python3 tools/model_converter/onnx_edit.py
--input_model=${MODEL_FOLDER}/yolo-v3-tiny.onnx
--output_model=${MODEL_FOLDER}/yolo-v3-tiny-nchw.onnx