CODEWITHSUNDEEP
computer-visiondeep-learningconv-neural-networkdlib
StereoMatching
StereoMatching

Reputation: 5019

How to read the data of Adience benchmark(data set for gender and age classification)?

I am trying to train a gender and age classification by cnn, using the data at adience and I got two questions.

1 : According to the website, the bounding box of the faces are recorded in the fields "x,y,dx,dy". As an example, from fold_frontal_0_data.txt,the first data is

image name  : 10424815813_e94629b1ec_o.jpg  
(x,y,dx,dy) : 301 105 640 641

However, the data of the bounding box is weird, because the size of the image is 600x601 only, no matter I treat (x,y,dx,dy) as (left, top, width, height) or (left, top, right, bottom), I cannot crop the face as expected. How could I crop the face properly?

2 : Do I need to do face alignment on training data? Or I only need to do face alignment when testing?Or both?Thanks

This is the image--10424815813_e94629b1ec_o.jpg 10424815813_e94629b1ec_o.jpg

Upvotes: 3

Views: 1252

Answers (1)

ilkyu tony lee
ilkyu tony lee

Reputation: 103

I am the other guy who dealing with Adience dataset.

File name of image you uploaded is

coarse_tilt_aligned_face.2.10424815813_e94629b1ec_o.jpg

not

image name : 10424815813_e94629b1ec_o.jpg

If you see README.txt precisely, you can find

original_image - the filename. Unfortunately we are unable to provide a URL, as we only have the file name and there is no way to reproduce it from the API. A link to the account is provided for accrediting.

(x,y,dx,dy) informations are should be gained from original_image.

Upvotes: 1

Related Questions