the params in main.py for save model

[nanhui69]

remember best Accu and save checkpoint

    is_best = accu > best_accu
    best_accu = max(accu, best_accu)

    if epoch in args.decay_epoch:

the is_best param is not use for save model anywhere , if i want to auto save model according the best metric ,how to define it??/

[chufengt]

if is_best: save_checkpoint(...)

[nanhui69]

if is_best: save_checkpoint(...)

ok， got it , thx doctor

[nanhui69]

another question is that does this model is perform better than resnet50 model such as this repo ?? furthermore ,what‘s mean for this li for idx in range(105): dataset['att_name'].append(data['peta'][0][0][1][idx, 0][0])
in the transform_peta.py，isn't attribute num 65 for peta data？？？

[chufengt]

What do you mean? There is no transform_peta.py.

[nanhui69]

What do you mean? There is no transform_peta.py.

just refer this train code :—> https://github.com/dangweili/pedestrian-attribute-recognition-pytorch ,and i find you make an ccommment under this repo. i cannot understand” for idx in range(105): dataset['att_name'].append(data['peta'][0][0][1][idx, 0][0])　“ in transform_peta.py

[chufengt]

It seems PETA originally have 105 attributes, including many color attributes, but only 61 attributes are commonly used. I didn't use this script. I downloaded the train/test list from https://github.com/asc-kit/vespa/tree/master/generated .