Create a handler for animal based images

rohanakut commented 2 years ago

This issue deals with creating handlers for images containing animals. The handler should work on images like screenshot (Source - https://lacombeonline.com/articles/saving-grace-animal-society-to-host-pints-for-pets-in-lacombe-this-weekend)

The main implementation tasks for this issue are:

[x] Determine whether the image belongs to animal category. This is done using the graphic tagger
[x] (Updated on Aug 1): Since the azure categories does not work well in detecting animals, we need to create a logic that bypasses the azure categoriser. This could be done using semantic segmentation and current object detection model.
[ ] Refer to the object detection model trained for animal detection. The model has been trained as mentioned in #455
[ ] Create a rendering with detected animal as the centre of focus.

rohanakut commented 2 years ago

Update: Similar to #469 the azure model does not detect animals very well. However, the semantic segmentation model and the object detection model do detect animals very well. Hence we need to create a custom check to bypass the azure categoriser for animal categories

rohanakut commented 2 years ago

@jeffbl this handler might not happen in the near future. I have the ML model to work with, but currently for my thesis I do not see a significant contribution of this handler.

rohanakut commented 2 years ago

@jeffbl as I mentioned earlier I do not plan to implement this for my thesis. Can this issue be closed?

jeffbl commented 2 years ago

Moving to backlog. @rohanakut Can you comment on what someone would need to do to move this forward, if you think it would be feasible?

Shared-Reality-Lab / IMAGE-server

Create a handler for animal based images #471