AndreySheka/kd_oe_hpe

Preparing

1. Download datasets 300W-LP, AFLW2000 [1]; AFLW [2]; BIWI [3] and extract its to data/. Folder should be looks like:

   data/
       300W_LP.json
       AFLW.json
       AFLW2000.json
       BIWI.json
       300W_LP/
           AFW/
               AFW_134212_1_0.jpg
               AFW_134212_1_0.mat
               ...
           ...
       AFLW/
           data/
               flickr/
                   0/
                       image00002.jpg
                       image00013.jpg
                       ...
                   2/
                   3/
           aflw.sqlite
       AFLW2000/
           image00002.jpg
           image00002.mat
           ...
       BIWI/
           hpdb/
               01/
                   frame_00003_rgb.png
                   frame_00003_pose.txt
                   ...
               02/
               ...
               24/
   

   [1] 300W-LP and AFLW2000 datasets: http://www.cbsr.ia.ac.cn/users/xiangyuzhu/projects/3DDFA/main.htm

   [2] AFLW dataset: https://www.tugraz.at/institute/icg/research/team-bischof/lrs/downloads/aflw/

   [3] BIWI Kinect Head Pose dataset: https://icu.ee.ethz.ch/research/datsets.html

2. Install python modules:

   pip3 install -r requirements.txt
   

Evaluation

Download the models from the link and place them in the models/ folder. Folder should be looks like:

models/
    ap1/
        resnet18.pth
        resnet34.pth
        resnet50.pth
        resnet101.pth
        resnet152.pth
    ap2/
        resnet18.pth
        resnet34.pth
        resnet50.pth
        resnet101.pth
        resnet152.pth
    ap3/
        resnet18.pth
        resnet34.pth
        resnet50.pth
        resnet101.pth
        resnet152.pth

For evaluation use script src/main.py:

python3 src/main.py --protocol 1 --arch resnet50

Available protocols:

• 1 : AFLW2000 and BIWI
• 2 : AFLW-test subset
• 3 : BIWI-test subset

Available architectures:

• resnet18
• resnet34
• resnet50
• resnet101
• resnet152

Bounding box labeling

Folder data/ contains couple of JSON files with bounding box labeling. These files have following format:

{
    "filename": [
        {"bbox": [x_min, y_min, x_max, y_max], "type"?: "train|test", ...},
        ...
    ]
}

Citing

TBD