def load_pretrained()

in utils/util.py [0:0]

• 15 lines of code
• 5 McCabe index (conditional complexity)

def load_pretrained(args, model, logger=print):
    ckpt = torch.load(args.pretrained_model, map_location='cpu')
    if len(ckpt) == 3:  # moco initialization
        ckpt = {k[17:]: v for k, v in ckpt['state_dict'].items() if k.startswith('module.encoder_q')}
        for fc in ('fc_inter', 'fc_intra', 'fc_order', 'fc_tsn'):
            ckpt[fc + '.0.weight'] = ckpt['fc.0.weight']
            ckpt[fc + '.0.bias'] = ckpt['fc.0.bias']
            ckpt[fc + '.2.weight'] = ckpt['fc.2.weight']
            ckpt[fc + '.2.bias'] = ckpt['fc.2.bias']
    else:
        ckpt = ckpt['model']
    [misskeys, unexpkeys] = model.load_state_dict(ckpt, strict=False)
    logger('Missing keys: {}'.format(misskeys))
    logger('Unexpect keys: {}'.format(unexpkeys))
    logger("==> loaded checkpoint '{}'".format(args.pretrained_model))