These are chat archives for beniz/deepdetect

Nov 2016
Nov 22 2016 10:44

Hello, I asked a question on github, about installing deepdetect. So, I dowloaded docker set it up, etc and then did

  1. docker pull beniz/deepdetect_cpu
  2. docker run -d -p 8080:8080 beniz/deepdetect_cpu
  3. curl http://localhost:8080/info (got proper results)
  4. curl -X PUT "http://localhost:8080/services/imageserv" -d ... (got proper results)
  5. curl -X POST "http://localhost:8080/predict" -d ... (got proper results).

So, now my question is, is deepdetect properly installed? I for example, I want to run it with an image img.jpeg on my desktop. How do I go about that?

p.s. sorry for opening an issue on github, I was not aware of this chat room

I think the next step I need to do is:

cd cpu
docker build -t beniz/deepdetect_cpu --no-cache .

but when I do cd cpu, I get "no such file or directory"

Emmanuel Benazera
Nov 22 2016 10:58
@gtsoumis hi, not sure I understand what your issue is if prediction works as expected... Are you familiar with docker ? If not, please read on docker documentation from docker, we can't help you much with these basics. If it is not docker related, try to explain what you want to do, like predict on your own image ? If the latter, just modify the POST call as needed.
Nov 22 2016 11:34

I just rank

curl -X POST "http://localhost:8080/predict" -d "{\"service\":\"imageserv\",\"parameters\":{\"input\":{\"width\":224,\"height\":224},\"output\":{\"best\":3},\"mllib\":{\"gpu\":false}},\"data\":[\"$_35.JPG?set_id=2\"]}"

and got

{"status":{"code":200,"msg":"OK"},"head":{"method":"/predict","service":"imageserv","time":1915.0},"body":{"predictions":[{"uri":"","classes":[{"prob":0.4729715883731842,"cat":"n03314780 face powder"},{"prob":0.2600928246974945,"cat":"n02840245 binder, ring-binder"},{"last":true,"prob":0.029175568372011186,"cat":"n03961711 plate rack"}]}]}}

too much cat and no cellphone
Emmanuel Benazera
Nov 22 2016 11:52
You are using an Imagenet pre-trained model as it seems. You'd need to build your own model in order to get proper results for your application at hand.
FYI, here is the output of an inception_v4 net pre-trained on Imagenet: {"status":{"code":200,"msg":"OK"},"head":{"method":"/predict","service":"imageserv","time":434.0},"body":{"predictions":[{"uri":"","classes":[{"prob":0.4182678759098053,"cat":"n03180011 desktop computer"},{"prob":0.11038176715373993,"cat":"n04152593 screen, CRT screen"},{"prob":0.08769628405570984,"last":true,"cat":"n03642806 laptop, laptop computer"}]}]}}