I have had a go of the API and trained a basic model. It is pretty good and easy to use. I have a use case and not sure of the best approach or if it is feasible. Or if anyone has done anything similar?
Say we have a number of fixed cameras getting about 30,000 images per month (in a different place each month), The camera is set off by motion, we are only interested in images of certain animals, we want to ignore anything set off by a car, a person or just the wind (ie no new objects in the image).
What i am thinking might be an approach, rather than trying to train a model for all sorts of different animals we might try and identify images we are not interested in, therefore reducing the number of images a person has to look through.
1st step would be to identify all images which have something in them different to the empty scene
Then discard anything with a person in it
Then discard anything with a vehicle in it
Then discard anything with a cow in it (we are not interested in cows)
Then whatever is left over could be of interest and would be looked at by someone.
I thought this might be a better approach than trying to train a model to recognize a dog, cat, deer, pig, bird, rabbit, reptile etc etc. especially given that they are not usually very clear.
Is this feasible on the clarifai platform? Another thing to complicate it further is a lot of it is night vision..