Yikes ... Big Brother is Watching (aka GoogleCloudService)

Found 20 faces
Writing to file faces.jpg.out.jpg
Labels for image faces.jpg:
person (score: 0.966)
facial expression (score: 0.906)
painting (score: 0.756)
profession (score: 0.743)
portrait (score: 0.699)
emotion (score: 0.680)
moustache (score: 0.512)
 
 
Labels for image kitchen.jpg:
kitchen (score: 0.983)
home (score: 0.951)
cabinetry (score: 0.911)
room (score: 0.911)
residential area (score: 0.879)
estate (score: 0.878)
property (score: 0.823)
dining room (score: 0.806)
real estate (score: 0.706)
flooring (score: 0.695)
wood flooring (score: 0.693)
hardwood (score: 0.684)
interior design (score: 0.559)
floor plan (score: 0.522)
 
Labels for image plumbing.jpg:
trigger (score: 0.857)
exhaust system (score: 0.517)
 
 
Labels for image ship.jpg:
vehicle (score: 0.949)
ship (score: 0.912)
ship of the line (score: 0.875)
ocean liner (score: 0.813)
warship (score: 0.779)
sailing ship (score: 0.778)
frigate (score: 0.776)
full rigged ship (score: 0.731)
naval ship (score: 0.725)
barque (score: 0.715)
mode of transport (score: 0.705)
watercraft (score: 0.657)
clipper (score: 0.646)
tall ship (score: 0.556)
caravel (score: 0.548)

c.m.service.GoogleCloudService - 9022 total ms

 

Above is the results of using the new GoogleCloud  Service ...   I believe someone was interested in this and started trying to interface it with Python.  The client credential part was a pain .. but in the end these are the results.  Total processing took 9 seconds.  There are a few more examples I mean to look at "Land Mark Detection" & "Text Detection"  .. which I'll add shortly..

Some of the requirements of a authorization is submitting a credit card.  I don't remember if there is a free quota or not, but this has changed.  In the past Google used to allow free access without having to submit a credit card as long as you stayed under a certain quota ...   Times have changed ;)

Incapsulated in a service, it should allow easy access and interfacing with other services.

References :


Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
jaideepjoshi's picture

Google Vision API to detect and label objects

I was thinking about the staatement that Gael has made more than once. That ideally, one needs a mechanism whereby all inmoovs need to, as an example, detect a green ball. 

While I did some haar cascade training, using OpenCV for a banana, I have no means of doing it for all the known objects. Also, it took 2 days to do just one object and generate a xml file. So, in the absence of a OpenSource service, a service such as this could be used, as a starter. I believe it is 1000 free lookups per month.

The flow I was trying to generate was for inmoov to take a picture (single frame image) of an object I am holding, do a label lookup in the background and use Acapela to tell me the attributes of the image when received from the Vison Api. The speed is dependent on the API/internet connectivity itself. I have parts of all this working but dunno how to do this from MRL.

I have uploaded the basic python scipts here

http://myrobotlab.org/content/executing-python-script-mrl

Thanks

 

GroG's picture

It works as I would expect it

It works as I would expect it .. basically similar to reverse search on Google Image search..

It got Ball ... not "Green Ball" .. but Ball ;)

Labels for image greenball.jpg:
ball (score: 0.7874977)
earth (score: 0.78296065)
world (score: 0.78016394)
sports equipment (score: 0.71067965)
sphere (score: 0.68601936)
shape (score: 0.5628206)
globe (score: 0.54525816)
football (score: 0.52819556)
 
I've cleaned up credentials, and some internal methods...
But there is still more work to be done..
 
I've got to :
  • clean up dependencies 
  • make ivy file
  • make service page
  • make example python usage

 

 

jaideepjoshi's picture

Maybe Image Features will return a color.

Thanks for all your efforts. I think this could be a good addition to MRL.