mirror of
https://github.com/blakeblackshear/frigate.git
synced 2026-05-01 11:07:41 +03:00
docs tweak
This commit is contained in:
parent
914941090b
commit
3201985359
@ -171,7 +171,7 @@ When choosing images to include in the face training set it is recommended to al
|
||||
- If it is difficult to make out details in a persons face it will not be helpful in training.
|
||||
- Avoid images with extreme under/over-exposure.
|
||||
- Avoid blurry / pixelated images.
|
||||
- Avoid training on infrared (gray-scale). The models are trained on color images and will be able to extract features from gray-scale images.
|
||||
- Avoid training on infrared (gray-scale). The models are trained on color images and will not be able to extract features from gray-scale images.
|
||||
- Using images of people wearing hats / sunglasses may confuse the model.
|
||||
- Do not upload too many similar images at the same time, it is recommended to train no more than 4-6 similar images for each person to avoid over-fitting.
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user