RGB or BGR input?

[acnazarejr]

The readme.md has a code example of how to use the code:

img = cv2.imread("ivan.jpg")
detector = MTCNN()
print(detector.detect_faces(img))

In this example, the img is a BGR image (opencv format). But, I think the correct is to convert the image to RGB.

I did some tests, and the model has better results with RGB images.

[Andriy963]

I tested this solution on Wider Face validation dataset for both cases(RGB and BGR) and got next results:
For BGR images(using code from readme file at this repo):
==================== Results ====================
Easy Val AP: 0.5647998370785989
Medium Val AP: 0.5219529625700366
Hard Val AP: 0.2856212575529135

For RGB images(with converting to RGB after image reading):
==================== Results ====================
Easy Val AP: 0.8069582917049369
Medium Val AP: 0.7637617089872778
Hard Val AP: 0.4867522878479363

So, there should be converting to RGB format before processing.

[nyck33]

@Andriy963 @acnazarejr Great to know, I just posted on StackExchange about this how pyplot.imread() was giving better results than the cv2.imread()

However, now my output when I cv2.imwrite() looks very blue.

[raviam]

@nyck33 For the blue output images, you can change the colormap of the image before cv2.imwrite() by doing image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR) and passing it to cv2.imwrite()

[punitha-valli]

@acnazarejr @Andriy963 @nyck33 @raviam @akofman

Can you please tell me, which script is suitable to get the boundary box details as ".data" format, for my dataset (OULU),

Thanks in advance