Giters

openvinotoolkit / training_extensions

Train, Evaluate, Optimize, Deploy Computer Vision Models via OpenVINO™

Home Page:https://openvinotoolkit.github.io/training_extensions/

Repository from Github https://github.comopenvinotoolkit/training_extensionsRepository from Github https://github.comopenvinotoolkit/training_extensions

ONNX dynamic shaping error in OpenCV embedding shape interpolate

eugene123tw opened this issue · comments

commented

Describe the bug
ONNX dynamic shaping not working with Model API inference_adapter.embed_preprocessing.

[CPU] Interpolate node with name 'Interpolate_65093' Check 'cmp::le(m_min, u) && cmp::le(u, m_max)' failed at src/core/shape_inference/include/shape_infer_type_utils.hpp:54:
Value -1 not in range [0:18446744073709551615]

Steps to Reproduce

  1. otx train --config src/otx/recipe/detection/atss_mobilenetv2.yaml --data_root tests/assets/car_tree_bug --work_dir otx-workspace/atss_mobilenetv2
  2. otx export --work_dir otx-workspace/atss_mobilenetv2 --export_format ONNX
  3. otx test --work_dir otx-workspace/atss_mobilenetv2 --checkpoint otx-workspace/atss_mobilenetv2/.latest/export/exported_model.onnx

Environment:

  • OS:
  • Framework version:
  • Python version:
  • OpenVINO version:
  • CUDA/cuDNN version:
  • GPU model and memory:
commented

@eugene123tw feel free to remove all the related dynamic h w setting in onnx configurations: detection, instance segmentation and keypoint detection are affected