new test_example.py.

[jerryum]

first of all, thank you to upload test_example.py in PromptPAR

When I tested the model using test_example.py, I learned several things for the model.

• I see 105 attributes for each image in PETA from the PETA.mat file but the model is limited to use the attribute only 35. Is it right understanding? this is a part of your model.py. If my understanding is correct, any reason to limit down to use 35 only out of 105?

datasets_attrnum={'PA100k':26,'RAPV1':51,'PETA':35,'PETAzs':35,'UPAR':40,'RAPzs':53,'RAPV2':54,'WIDER':14,"RAPV1Expand":51}

from scipy.io import loadmat import os save_dir = '/home/jerryum/works/OpenPAR/dataset/PETA/' peta_data = loadmat(os.path.join(save_dir, 'PETA.mat')) raw_attr_name = [i[0][0] for i in peta_data['peta'][0][0][1]] raw_label = peta_data['peta'][0][0][0][:, 4:]

print (len(raw_attr_name)) print (raw_label.shape) 105 (19000, 105)

• Since model learns only 35 attribute, your new test_example.py throws an error - since the sample attribute list in the code has 50 attribute.

  
  File [~/works/OpenPAR/PromptPAR/test_example.py:132](https://vscode-remote+wsl-002bubuntu.vscode-resource.vscode-cdn.net/home/jerryum/works/OpenPAR/~/works/OpenPAR/PromptPAR/test_example.py:132), in validator(model, valid_loader, ViT_model, criterion, args)
  [130](https://vscode-remote+wsl-002bubuntu.vscode-resource.vscode-cdn.net/home/jerryum/works/OpenPAR/~/works/OpenPAR/PromptPAR/test_example.py:130) for pidx in range(len(image_names)):
  [131](https://vscode-remote+wsl-002bubuntu.vscode-resource.vscode-cdn.net/home/jerryum/works/OpenPAR/~/works/OpenPAR/PromptPAR/test_example.py:131)     for aidx in range(len(attributes)):
  --> [132](https://vscode-remote+wsl-002bubuntu.vscode-resource.vscode-cdn.net/home/jerryum/works/OpenPAR/~/works/OpenPAR/PromptPAR/test_example.py:132)         if valid_probs[pidx][aidx] : 
  [133](https://vscode-remote+wsl-002bubuntu.vscode-resource.vscode-cdn.net/home/jerryum/works/OpenPAR/~/works/OpenPAR/PromptPAR/test_example.py:133)             pred_attrs[pidx].append(attributes[aidx])
  [134](https://vscode-remote+wsl-002bubuntu.vscode-resource.vscode-cdn.net/home/jerryum/works/OpenPAR/~/works/OpenPAR/PromptPAR/test_example.py:134)         if valid_gt[pidx][aidx] : 

IndexError: index 35 is out of bounds for axis 0 with size 35

[jerryum]

I changed the attributes like below - same as the list from peta_pad.py and got a successful outcome.. Also, had to deal with several minor issues... FYI.

attributes = [
    'A pedestrian wearing a hat', 'A pedestrian wearing a muffler', 'A pedestrian with no headwear', 'A pedestrian wearing sunglasses', 'A pedestrian with long hair',
    'A pedestrian in casual upper wear', 'A pedestrian in formal upper wear', 'A pedestrian in a jacket', 'A pedestrian in upper wear with a logo', 'A pedestrian in plaid upper wear',
    'A pedestrian in a short-sleeved top', 'A pedestrian in upper wear with thin stripes', 'A pedestrian in a t-shirt', 'A pedestrian in other upper wear', 'A pedestrian in upper wear with a V-neck',
    'A pedestrian in casual lower wear', 'A pedestrian in formal lower wear', 'A pedestrian in jeans', 'A pedestrian in shorts', 'A pedestrian in a short skirt', 'A pedestrian in trousers',
    'A pedestrian in leather shoes', 'A pedestrian in sandals', 'A pedestrian in other types of shoes', 'A pedestrian in sneakers',
    'A pedestrian with a backpack', 'A pedestrian with other types of attachments', 'A pedestrian with a messenger bag', 'A pedestrian with no attachments', 'A pedestrian with plastic bags',
    'A pedestrian under the age of 30', 'A pedestrian between the ages of 30 and 45', 'A pedestrian between the ages of 45 and 60', 'A pedestrian over the age of 60',
    'A male pedestrian'
]

[1125178969]

The reason why only 35 of PETA's 105 attributes are chosen is because the other attributes have lower positive ratios, where we follow the settings given in the PETA paper

[1125178969]

Thanks for the correction.

[jerryum]

Thanks for the confirmation! By the way, the project is extremely helpful for me to understand the algorithm of the PAR....