by Andrew Grimes and Jim Garrett of Red Hat
Containerized app that serves a containerized Resnet18 deep learning image classification model using FastAPI. We used an ImageNet pretrained model that can predict 1000 different classes of general objects, the samples are animals but it will work with anything. See class list here.
ResNet18 https://www.mathworks.com/help/deeplearning/ref/resnet18.html
Sample Backend App and Input Image:
Sample Text Output: {"success":true,"predictions":[{"label":"black-and-tan coonhound","probability":0.5641617774963379},{"label":"Doberman","probability":0.3869141638278961},{"label":"bluetick","probability":0.012455757707357407},{"label":"Rottweiler","probability":0.007904204539954662},{"label":"Gordon setter","probability":0.006333122029900551}]}%
OpenShift/ROSA instructions (deck to be created)
-
Bring up a generic ARO cluster in Azure. This also works with OpenShift or other Managed OpenShift Options to demonstrate portability.
-
Create a Project and use the Git build option taking the defaults. watch the build logs until complete to see how ROSA builds. This Repo: https://github.com/emcon33/inference-rosa-workshop
-
Obtain your URL for the web front end from the image via topology map, it should open to a place holder page.
-
From local system use a test1.image with your URL updated to this command and "curl" your image to the API front end. curl -X POST -F image=@test3.jpeg "https://inference-aro-workshop-test7.apps.aro-vs2cl.zpq2.p1.openshiftapps.com/api/predict"
-
Additional test images are available on the source git repo or use your own, it works with animals and objects.
Optional front end to this project. Here is the git repo just follow the instructions to fork and modify the URL from this deployed repo.
https://github.com/emcon33/inference-aro-frontend
Original source https://towardsai.net/p/machine-learning/build-and-deploy-custom-docker-images-for-object-recognition
Forked from this github https://github.com/hasibzunair/imagercg-waiter
`