Question: Can we make detection area bigger?

[masterofobzene]

Refering to output faces.
Images are better than words.

From this:

To this:

[freearhey]

I have just added a new --scale argument through which you can control the size of the detection zone. For example:

python extract.py --input examples --scale 1.2

I hope this helps.

[masterofobzene]

I have just added a new --scale argument through which you can control the size of the detection zone. For example:

python extract.py --input examples --scale 1.2

I hope this helps.

A little problem I see is that when the scale increases, the detection gets really worse. Is there any way to fix this?

[freearhey]

Could you provide any example?

[masterofobzene]

using -s 1.6 (increasing this value makes it worse)
all have clear faces
images res +- 1000x800

Value at 1.2:

Value at 0.8:

[freearhey]

Fixed in #4

[masterofobzene]

Hi there,
It seems the area is bigger now but I'm getting a lot of black bars in the output images, like if the resolution is fixed but the image inside is not fullsize.

Example:

(blurred for privacy)

[freearhey]

This is expected behavior since all found faces are now placed in the middle of the frame.

[masterofobzene]

Not a very good behavior IMO why would you leave black bars on the images?

[freearhey]

Because, as far as I know, this is the only way to put the face in the center of the frame regardless of its position in the original photo.