Updated the blurring pipeline
Updated the blurring pipeline
There are now 3 blurring strategies : FAST
, COMPROMISE
and QUALITATIVE
, each with different quality and speed.
There is also a new entry point : redo-sequences
, allowing a sequence of pictures to be processed again.
These strategies employ semantic segmentation and object detection AIs to find the regions of a picture to blur within reasonable time.
More can be read in server/readme
and server/src/blur/readme
.
I did not run CD/CI, the docker installation is working fine but I can't get tests to work.
Update notes
Old installations must replace there ENABLE_BLUR
environment variable with the new BLUR_STRATEGY
one.
The algorithms used have quite a lot of dependencies, making python 3.9 required, 3.10+ is not yet supported by the YOLOv6 dependency.
There is one dependency (pytorch) which installation depends on the OS running the docker container, the Dockerfile contains the command for Linux, it will probably cause problem on Windows.
Sequences that were previously processed (blurred) are totally compatible but can be re-processed using redo-sequences
.
The docker installation has been modified to reflect these changes.