Object detection
Opened this issue · 2 comments
My apologies in advance for abusing the issues to ask a rather naive question.
I'm trying to gauge how far off some reasonably simple object detection would be in grenade? (Perhaps something in a similar vein to the tensorflow object detection api or YOLO)
Interesting question. I'll speculate as I have not done this to date.
Comparing grenade to Darknet is a fair thing to do I think. Specifying the network would be pretty trivial. The only exception is that I don't actually know what his region layer actually does; but I can't imagine it being too hard to port over.
After that it comes down to two things really: getting data in and out; and speed. Grenade doesn't implement any parsers or data formats, it just takes a Vector, so you'll have to write some code or use another Haskell library to handle them; for speed, although it's CPU bound at the moment, it's pretty quick for convolution tasks.
I think Grenade could be applied.
I may give it a try then, thanks. Will report back if I decide this is the way to go.