fdbtrs / PocketNet

Official repository for PocketNet: Extreme Lightweight Face Recognition Network using Neural Architecture Search and Multi-Step Knowledge Distillation
Other
56 stars 10 forks source link

where the pre-trained teacher model? #6

Open xiakj opened 2 years ago

xiakj commented 2 years ago

`backbone_teacher = iresnet100(num_features=cfg.embedding_size).to(local_rank) try: backbone_teacher_pth = os.path.join(cfg.teacher_pth, str((epoch + 1)*11372) + "backbone.pth") backbone_teacher.load_state_dict(torch.load(backbone_teacher_pth, map_location=torch.device(local_rank)))

        if rank == 0:
            logging.info("backbone teacher loaded for epoch {} successfully!".format(epoch))
    except (FileNotFoundError, KeyError, IndexError, RuntimeError):
        logging.info("load teacher backbone for epoch {} init, failed!".format(epoch))
        break`

I can not find the teacher model.

fdbtrs commented 2 years ago

You can use the pretrained model from ElasticFace

soulyiran commented 2 years ago

You can use the pretrained model from ElasticFace @fdbtrs I have download the pretrained model from ElasticFace, but it seems like the embedding_size of these model are 512. Could you please provide the 128d version? or where could i find these pretrained model