Future feature ideas

[nathanielrindlaub]

Tracking low-priority feature ideas:

• Ability to validate/invalidate objects from pills in image table
• Order labels alphabetically in the labels filter
• Customizable “welcome” screen that reminds users what their job is and what they’re looking for & who to contact if they see a rat etc..
• Email UX improvements
  • bake bounding box into the embedded image in alert emails
  • buttons for validating/invalidating objects included in emails

[nathanielrindlaub]

• consider offering a global TZ selection (think AWS region selection drop down in AWS console) so that if a user is in Germany but wants to look at CA images that were taken between two dates, they can perform the query in the same timezone as the cameras.

[nathanielrindlaub]

• an option to blur humans in photos? We could just trust the output of MegaDetector and bake the blur into the image's pixels, or perhaps render the blur entirely on the front-end, which would allow us to do things like only show the blur if the user is a volunteer labeler (this feature was inspired by this repo here)

[nathanielrindlaub]

Non-ML automated image processing options to consider:

• option to average ML prediction confidences across sequences/bursts of images (would depend on https://github.com/tnc-ca-geo/animl-api/issues/77)
• option to disregard objects after object detection step that are too big or too small (object's bounding box takes up too much or too little of the frame)

[nathanielrindlaub]

• auto confidence threshold adjusting: we would have to store all of the predictions models return, not just the ones over the confidence threshold, and then periodically run some code that would assess whether increasing or decreasing the confidence threshold would reduce false negatives or false positives. It's also worth setting the confidence threshold on a per-camera basis. Would be challenging but cool!