Fork me on GitHub
#data-science
<
2019-10-22
>
blueberry12:10:04

New book release 0.11.0 sheepy sheepy sheepy Adaptive Learning Rates: RMSprop and Adam https://aiprobook.com/deep-learning-for-programmers/

sheepy 6
🚀 2
fabrao17:10:58

hello all, anyone used GraalVM with libs from Python?

fabrao18:10:27

@metasoarous I´ve never heard about it. Do you think can I use the opencv stuffs with this?

metasoarous18:10:59

I imagine. This is @chris441's baby; He may know whether anyone has tried opencv with it yet.

chrisn18:10:52

@fabrao Likely, yes. I know that someone has run the libpython-clj unit tests from graalVM. We have used matplotlib which I can't imagine is any crazier than opencv.

gigasquid18:10:30

if you want to use opencv you can also check out https://github.com/hellonico/origami

chrisn18:10:05

Also: https://github.com/techascent/tech.opencv And I use opencv to validate tvm-clj.

fabrao18:10:57

@gigasquid I´ve used it, but other libs like face-recognition only in python

👍 1
chrisn18:10:22

@fabrao - Are you using MTCNN?

fabrao18:10:06

I tried many libs, but the https://github.com/ageitgey/face_recognition is the simplest lib I´ve used, you can get recognition with one picture

fabrao18:10:04

even if you rotate head a little

chrisn18:10:52

Cool 🙂. I did facial rec for a while. @gigasquid told me about insightface which I really liked for created the actual feature vectors.

fabrao18:10:02

Have you used it with core.async?

fabrao18:10:28

I´m thinking using it like chan of images to speed up processing and skipping some frames to increase fps

chrisn18:10:39

I have not used it with core.async. When I was working with that stuff I found batching to be the most effective. Processing 32 image batches or something (whatever your RAM allows) worked very well as opposed to processing 1 image.

chrisn18:10:54

But I was finding a lot of faces in 1 image.

chrisn18:10:04

So I had a lot of faces to create vectors out of.

chrisn18:10:48

Parallelizing the actual find-faces pathway was also useful. But for our use case MTCNN at that time was markedly better than dlib. Times change, however, as does the version of dlib so YMMV.

fabrao19:10:50

Did you use it with clojure or python?

fabrao19:10:21

I need it with face recognition, not only detection

chrisn19:10:42

We used both mtcnn and insightface from python.

chrisn19:10:58

via clojure.java.shell

chrisn19:10:56

There are better face detectors than MTCNN now though. I think there are a few of them in the deepinsight repo that contains insightface: https://github.com/deepinsight

chrisn19:10:26

This version of MTCNN worked the best: https://github.com/ipazc/mtcnn

chrisn19:10:16

OK, I have to stop 🙂. I could go on quite a bit. Good luck!