In the FOMO blog how do we map the 12x12 output back to 96x96 input

Question/Issue: The FOMO output is 1/8 of the input resolution. What is the post processing steps need to be taken so that the 1/8th resolution output is mapped to input resolution. Can you please provide how in FOMO model which you create does it.

**Project ID:**Object Detection

Context/Use case:

Hi @amitroy

Right now we have post processing via performance calibration in our enterprise tier:

The post-processing offered includes:

  1. Averaging Scores Over a Window: Before any decisions are made, the model’s output scores are averaged over a specified duration to smooth out any abrupt fluctuations.

  2. Applying a Threshold: Only the top score, after averaging, is considered. If this score surpasses a predetermined threshold, it indicates the presence of a detected event.

  3. Suppression Period: After a positive event detection, there’s a period where any subsequent detections are temporarily ignored. This avoids rapid repeated detections of the same event and reduces false positives.

You should be able to perform more custom post processing in a notebook via our python SDK, although this is something that we could add to the discussion point for when you get time to have a meeting with our sales team as there are teams that can hopefully help work with you to explore this need at a deeper level.