tusing / nms-speedup Goto Github PK

It would be nice to have a simple python GUI for viewing the images (when they are available) and viewing the bounding boxes before and after. This can be very simple and should probably be implemented in tests.py

Hello.

I am trying to this nms code from this repo.

However, I couldn't get forward for the nms.so file.

Could you give me the way?

Currently, nmsGPU.cpp is compiled separately, and the Makefile does not incorporate it. There are several options

• Convert nmsModule.c to .cpp, alter Makefile accordingly to link with nmsGPU.cpp. Should be easiest
• Convert nmsGPU.cpp to .c, alter Makefile accordingly
• Leave nmsGPU as cpp, nmsModule as .c, link them somehow???

The file tree is really messy right now. We should clean it up

• dataset/
  • Some way of organizing our datasets here
• src/
  • .py, .c, .cl, etc.
• Makefile
• README.md

Our scalar code is too fast. The SIMD and OpenMP implementations are not significantly faster, even though avx performs theoretically 8x faster.

This is likely because non scalar algorithms generally perform more comparisons than necessary. This is something we will have to deal with in the GPU version as well.