Dear Xilinx Support Team,
I am currently experiencing issues with running the SuperPoint model from the Vitis AI Model Zoo (version 3.0) within the Vitis AI 3.0 GPU Docker container. Despite following the installation and setup instructions closely, including using TensorFlow-GPU version 1.15.0, matching the CUDA device version as recommended, and setting up the environment as per the guidelines, I came across several problems when attempting to run the run_demo.sh script.
The repo I am trying to run: https://github.com/Xilinx/Vitis-AI/blob/3.0/model_zoo/model-list/tf_superpoint_mixed_480_640_52.4G_3...
These steps worked to my knowledge and built/ran docker container: https://xilinx.github.io/Vitis-AI/3.0/html/docs/install/install.html#option-2-build-the-docker-conta...
Firstly, the script fails with errors related to GPU device assignment, specifically indicating that no CUDA-capable device is detected even when a GPU is present and supposedly configured correctly (CUDA and GPU Detection Issues). Additionally, when I bypass GPU usage, it results in errors due to the absence of a GPU, as the TensorFlow graph is set to utilize GPU:0.
I tried to migrate to TensorFlow 2.x but I wasn't successful...I assume this is because there are significant differences between TensorFlow 1.x and 2.x, which led to compatibility issues and further errors.
Furthermore, I encountered numerous error messages related to versioning issues, TensorFlow dependencies (deprecated TensorFlow functions and attributes), and for example, encountered bugs such as "TypeError: load() missing 1 required positional argument: 'Loader'" due to incorrect usage of the YAML loader function, requiring specification of the loader type to resolve.
Given these challenges, I kindly request assistance in resolving these kind of issues to successfully run the SuperPoint demo within the Vitis AI environment. Any guidance on troubleshooting or steps to ensure compatibility and correct setup would be greatly appreciated.
Thank you for your support.
