[Experiment] Collect 4K crop img around detection bounding box #311
Description
Pitch
Currently, in order to save bandwidth, images sent to users (mainly firefighters) are compressed, even though they originate from 4K images.
In order to analyse alerts, users often need to zoom in on images, and detected smoke may appear pixelated at a certain zoom level.
It would be extremely useful to be able to make the best operational decisions and respond more quickly to a fire, and to have a better quality image in and around the detection area.
- [DRAFT] Improving image quality in/around the bbox - mix jpeg & 4K - Tiled rendering/DeepZoom #305 aims to try to see if merging 4K crop image and full image resized and compressed is relevant.
- Furthermore, it is not impossible that, on the front end, we may want to display the complete compressed image and, for example, have the 4K cropped images appear when the mouse hovers over them (see screenshot illustration below)
We would therefore like to collect 4K cropped images around the detection area to verify that this higher-quality image helps with smoke analysis.
Depending on the display solution chosen, the developments made in this issue will be used to collect cropped images over the long term, to assist firefighters in their analysis of detections.
What should be done
- A function/class that, from 4K image captured, extract a crop image around the detection bounding box (bbow): bbox enlargement factor as a parameter (extract an image 20% larger than the bbox)
- For each detection sent to the API, the cropped image should be send to an S3 bucket -> (all detections or only a fraction?) -> To be discussed
S3 bucket info can be provided (but for developing the feature, might be relevant to use the local S3 service from pyro-envdev)
More information to be provided (ex where to add this feature in the detection process)
Happy to discuss it ! :)