DarkHelp VS pydensecrf

The code is open-source. The project starts with this: https://github.com/stephanecharette/DarkHelp/blob/master/src-apps/video_object_counter.cpp

In my repo (https://github.com/stephanecharette/DarkHelp) I say that? Please show me where.

From your description: I'd use the MoveDetect library to find where movement occurs, and then feed just those frames or videos into DarkHelp. Note that DarkHelp also has much richer JSON output than the vanilla Darknet CLI.

Came here to say the same. I work mostly on embedded devices, OpenCV, ML/CV type of products. And all our cmake files use globbing, and we love it. Here is an example CMake file for an open source tool I publish on github: https://github.com/stephanecharette/DarkHelp/blob/master/src-lib/CMakeLists.txt

I use the darknet C++ API, DarkHelp. But to be fair and in the spirit of full disclosure...I wrote it. :)

(I'm the author of DarkHelp, the C++ layer for Darknet.)

DarkHelp will load videos, run inference, and create a new output video. https://github.com/stephanecharette/DarkHelp

- PyDenseCRF : It does not have learnable parameters.

Thanks to your observation I was able to find this solution which resolved the issue.

The documentation for pydensecrf says it requires a recent version of Cython. Do you have a recent version of Cython installed, and is it set up enough that you can do, for instance, from Cython.Build import cythonize?