Properties of images for the best result

[akhilgupta0221]

• face_recognition version:
• Python version:
• Operating System:

Description

Using images to train the Model with face_recognition

Query

What are all the similar properties (i.e : Image size, resolution) all the image should have So, that face_recognition gives the best results.

[ageitgey]

Basically, the bigger the face area in the image, the better. So if each face area within the image is a few hundred pixels square, you will get the best results. Smaller faces have less detail and will be less accurate.

The size of the total image itself doesn't matter much for accuracy (since just the face area is what matters), but larger image sizes will make the whole system run much more slowly.

[akhilgupta0221]

Thanks !!!!

[akeilox]

I have hundreds of face pictures in 120x160 pixel dimension, they are all portrait pictures (clear view of face with same background). Do you foresee any issues with face detection for 120x160 pixel images?

[panchami88]

Hi, I am working face recognisation, while upload my images it showing error like ValueError: operands could not be broadcast together with shapes (1,2) (128,) , how I have to rectify it

[julianadormon]

I have hundreds of face pictures in 120x160 pixel dimension, they are all portrait pictures (clear view of face with same background). Do you foresee any issues with face detection for 120x160 pixel images?

@akeilox did you ever test using those small images? I really would like some recommended sizes but cannot find any info. I'm sure bigger is more accurate but I am looking for the sweet-spot.

[akeilox]

@julianadormon I didnt get to it