NVIDIA-Optical-Character-Detection-and-Recognition on Xavier AGX

[Jkan_arnick]

Hi, We are working on Optical-Character-Detection-and-Recognition related development on Xavier AGX and for that we used the code as it is on the following link - GitHub - NVIDIA-AI-IOT/NVIDIA-Optical-Character-Detection-and-Recognition-Solution: This repository provides optical character detection and recognition solution optimized on Nvidia devices.

While testing with their doc.jpg, we are not getting the same output. Attached the input image, expected vs our result image. The performance of the Xavier also very slow (taking ~2sec for the detection).

Do we need to enable any additional GPU configuration on Xavier or Anything we missed it? This will be really helpful to us to move further.

Thanks in advance.

[Morganh]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

@Jkan_arnick

While testing with their doc.jpg, we are not getting the same output. Attached the input image, expected vs our result image. The performance of the Xavier also very slow (taking ~2sec for the detection).

For Xavier, please take care of its compute capability when build TensorRT OSS.
i.e.,

 Notes: You can use the helper script to compile TensorRT OSS.

You can build again with the correct compute capability.

For fps, please use non-vit version for both OCDNet and OCRNet.