How to create .lst file for this project

Hi @nttstar @yingfeng @yuzhichang @zhangxu19830126 @kernel8liang I want to use my own dataset for finetuning the pre-trained age-gender model(gamodel-r50). And I am creating a directory ‘all_ages’ in the same directory in which face2rec.py(src/data/) is. Then I have 100 sub-directories in this directory data corresponding to 0-99 ages, which further contains images in it accordingly. But face2rec.py function doesn’t contain code or make_list function (even if I pass --list True to parser). The other option I am left with is to use im2rec.py from mxnet

But the format of .lst file that im2rec.py generates doesn’t matches with the format that parse_lst_line (face_preprocess.py) function processes.

def parse_lst_line(line):
  vec = line.strip().split("\t")
  assert len(vec)>=3
  aligned = int(vec[0])
  image_path = vec[1]
  label = int(vec[2])
  bbox = None
  landmark = None
  #print(vec)
  if len(vec)>3:
    bbox = np.zeros( (4,), dtype=np.int32)
    for i in xrange(3,7):
      bbox[i-3] = int(float(vec[i]))
    landmark = None
    if len(vec)>7:
      _l = []
      for i in xrange(7,17):
        _l.append(float(vec[i]))
      landmark = np.array(_l).reshape( (2,5) ).T
  #print(aligned)
  return image_path, label, bbox, landmark, aligned

It seems like aligned accepts a number(probably 0 or 1). Hence, I guess the format of a line in .lst file will be:

aligned img_path age bbox_x1 bbox_y1 bbox_x2 bbox_y2 p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 Am I correct? If not, then please provide the right way to do it.

Also, after creating .lst file, should I use face2rec.py as it is to create .rec file

[nttstar]

1. Yes, the first column indicates whether face images in this lst were already aligned, it should be 0 or 1.
2. Reply to you later.
3. Bounding box and landmark locations are not required if you already aligned them.
4. Now it should be absolute path.

The contribution is very welcome. You can contact me via email~

[RamatovInomjon]

Hello guys @nttstar @xizi @eeric @prashantg445 @zhangxiaopang88 . Thanks! This repo is awsome! pretrained models for age gender trained using mobilenet or is there other backbone used? I got “Incompatible attr in node at 0-th output: expected [256], got [1024]” error while using pretrained weights to start training