This is a presentation project, exploring the usage of NVCaffe to run CTPN. It was created to make use of the latest features in NVCaffe to improve memory and inference performance of CTPN.

NVCaffe: Link to fork used in the project

CTPN: Link to fork used in the project

1. Open this notebook - ctpn_with_nvcaffe.ipynb
2. Click Open in Colab
3. Connect to a Runtime with GPU and execute.

Have fun!

• Comparision of inference runtime of baseline against NVCaffe TRT Layer
• Memory consumption with FP16
• CTPN with TensorRT and Triton Inference Server
• Inference Optimization with Python Multiprocessing and Asyncio