search

jakowenko / double-take

Unified UI and API for processing and training images for facial recognition.
https://hub.docker.com/r/jakowenko/double-take
MIT License
1.23k stars 98 forks source link

[FEAT] Allow marking invalid matches #204

Closed hipitihop closed 2 years ago

hipitihop commented 2 years ago

Is your feature request related to a problem? Please describe. At times I get a positive match, but it is incorrect.

Describe the solution you'd like I would like to be able to mark matches as incorrect, and in turn have this taken into consideration for training or future matching.

Additional context e.g. in this case, it is an invalid match. It should really be unknown. This may be related to another issue where someone mentioned that side on faces get matches when they should not

image

NickM-27 commented 2 years ago

The request makes sense, but this requires the training programs behind the scenes (Deepstack, Compreface, FaceBox) to support this, which I'm not sure which ones actually do if any

jakowenko commented 2 years ago

Unfortunately, as @NickM-27 said, there isn't a way to do this to my knowledge. If one of the detectors does support this via their API, feel free to open a feature request for it.

My suggestion would be to look at your training images for Tony and remove any that are low quality or do not have a clear view of the face. Take selfies from your phone if possible. I've found this method produces the best results.

Thanks for checking out this project!