Write an inference package to classify an image using the current model and gate scripts.

[karmatai]

Description

In this card we are going to creat a python package that loads the orieantation classifier and 6-class classifier, blank page filter to a single pipeline to be used to classify the images from orchestration worker.

Illustration:

Subtask

• Decode the image and handle exif rotation angle datagracefully.
• Add script to filter blank pages by impelementing a ink coverage/variance threshold.
• Apply preprocessing to center crop and load the tensor and save it for later use.
• Update the pipeline to forward the tensor to orieantation classifier.
• Update the pipeline to forward the upright images and rotate 180 degree and forward for the rotated images.
• load the same tensor and forward to 6-class classifer to get 6-class probilits and save it with the rotation angle.

Reviewer

• @eroux

[eroux]

yes that looks good! please keep the detected orientation in the final output too