RuntimeError: Error(s) in loading state_dict for InceptionNet

trying to run inference using simple script

import model as models
import torch
import logging
logging.basicConfig(level=logging.INFO)

attr_num = 51
model = models.__dict__['inception_iccv'](
    pretrained=True, num_classes=attr_num)
checkpoint = torch.load('rap_epoch_9.pth.tar')
model.load_state_dict(checkpoint['state_dict'])
logging.info("loaded model checkpoint")

got RuntimeError: Error(s) in loading state_dict for InceptionNet

[chufengt]

I think the pre-trained model is good enough to address this problem. If the problem still occurs, you can keep the attribute with the largest confidence only.

[chufengt]

With the prefix module_, it’s a nn.DataParallel model.

Converting the model to nn.DataParallel before loading can solve the problem: model = torch.nn.DataParallel(model).cuda()