Yolo darknet


#1

YOLO stands for “You only look once” is currently is state-of-the-art for real time object recognition. As compared other algorithm like R CNN , mask RCNN and other Computer vision methods it is very fast to detect multiple object objects in real time senario with high accuracy. And there is an implementation for YOLO by “pjreddie” in C and CUDA which called as darknet. There are implementation s in other languages like python, java as well. And I think Julia should also have it .There are relevant links below:
Pjreddie darknet : https://pjreddie.com/darknet/
Tensorflow implementation of darknet: https://github.com/thtrieu/darkflow?files=1
Research papers:
https://arxiv.org/abs/1506.02640


#2

What are you proposing or asking here? Are you going to provide such an implementation?


#3

I want to provide an implementation and Iam trying to code but since I am new to Julia , it will take time to get implemented .But in some sense I am proposing it.


#4

I’ve seen a few people on Slack who have worked on Julia implementations - unfortunately the discussion seems to have been swallowed by the limited Slack history! If memory serves @ianshmean was involved in those discussions!?

But in short I think there are implementations around.


#5

Thanks , I will try to contact him.


#6

I got tripped up trying to write the loss functions in a way appropriate for Julia and haven’t made much progress I’m afraid. Not because they’re that tricky, more that it’s hard to learn from other code implementations because they’re so highly dimensional for their own optimization.

I do want to get this sorted soon.

I’m also keen to find a way to go from a darknet trained yolov3 model to ONNX to Flux.


#7

Maybe start from a pytorch implementation would be more comfortable…