model_weight.pth about alignedreid-re-production-pytorch

[huanghoujing]

Hi, thanks for your attention. After training is over, you can add --only_test true to test the model, keeping other parameters the same as your training command. In this case, model_weight.pth is not needed and you don't need to specify --model_weight_file. The program will automatically use ckpt.pth in your exp_dir.

[weiyuxingchen]

why is the size of 'model_weight.pth' and 'ckpt.pth' different

[huanghoujing]

@weiyuxingchen ckpt.pth contains model weights and optimizer intermediate states; model_weight.pth only contains model weights. I upload model_weight.pth only to save disk. For testing your just-trained model, you can specify --only_test true and the code will use ckpt.pth in your exp_dir to run testing.

[chj1933]

@huanghoujing ，Hello， if I want to save the final model_weight.pth, how should I do?

[huanghoujing]

Hi, example to extract model weight from checkpoint file and save it to model_weight_file (to save disk space) is shown below.

ckpt_file = 'ckpt.pth'
model_weight_file = 'model_weight.pth'
map_location = lambda storage, loc: storage
ckpt = torch.load(ckpt_file, map_location=map_location)
model_weight = ckpt['state_dicts'][0]
torch.save(model_weight, model_weight_file)