Seeking Guidance on Model Compatibility between OpenVINO 2021 and 2023

[max-my]

Dear OpenVINO Team,
I am currently facing an issue while migrating my model from OpenVINO 2021 to OpenVINO 2023. The model works perfectly on the 2021 version, providing accurate inference results. However, when I tried to run it on the 2023 version, the inference results were completely unusable. I noticed that there are differences in the data structures used for inference between these two versions—one uses blob, while the other uses tensor.
I would like to know if the models that work on OpenVINO 2021 are no longer supported in OpenVINO 2023. If they are still supported, could you please provide some guidance or resources on how to properly adapt my model for the 2023 version? I have checked the release notes and documentation, but I am still unsure about the specific steps I need to take.
Thank you very much for your help. I look forward to your response.
Best regards

[ilya-lavrenov]

Hi,
Please, read https://docs.openvino.ai/2023.3/openvino_2_0_transition_guide.html

In general, it's expected that models produced with OpenVINO Dev tools of 2021.x version, should be supported by 2023 release.

To provide more details, we need a model to reproduce / small sample app how you use this model via new API (maybe you did it wrong).

[max-my]

Thank you for your response. My issue has been resolved. It turned out that the problem was due to the default NCHW layout used by OpenVINO, while OpenCV Mat follows the HWC layout. I managed to solve it by using the cv::dnn::blobFromImage method to handle the dimension conversion. Thank you for your support and guidance!