slip/slip.py

# Copyright 2021 Zilliz. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import sys
from pathlib import Path

import torch
from torchvision import transforms

from towhee import register
from towhee.operator.base import NNOperator, OperatorFlag
from towhee.types.arg import arg, to_image_color
from towhee.types.image_utils import from_pil, to_pil

from tokenizer import SimpleTokenizer

def get_model(model):
    if isinstance(model, torch.nn.DataParallel) \
      or isinstance(model, torch.nn.parallel.DistributedDataParallel):
        return model.module
    else:
        return model

@register(output_schema=['vec'])
class Slip(NNOperator)
    """
    SLIP multi-modal embedding operator
    """
    def __init__(self, model_name: str, modality: str):
        super().__init__()
        sys.path.append(str(Path(__file__).parent))
        self.tokenizer = SimpleTokenizer()
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.model.to(self.device)
        self.model.eval()

        self.tfms = transforms.Compose([
            transforms.Resize(224),
            transforms.CenterCrop(224),
            lambda x: x.convert('RGB'),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                 std=[0.229, 0.224, 0.225])
        ])

    def __call__(self, data):
        if self._modality == 'image':
             vec = self._inference_from_image(data)
         elif self._modality == 'text':
             vec = self._inference_from_text(data)
         else:
             raise ValueError("modality[{}] not implemented.".format(self._modality))
         return vec.detach().cpu().numpy().flatten()

    def _inference_from_text(self, text):
        texts = tokenizer(texts).cuda(non_blocking=True)
        texts = texts.view(-1, 77).contiguous()
        embedding = get_model(self.model).encode_text(texts)
        embedding = embedding / embedding.norm(dim=-1, keepdim=True)

    @arg(1, to_image_color('RGB'))
    def _inference_from_image(self, img):
        img = self._preprocess(img)
        img = img.to(self.device)
        embedding = get_model(self.model).encode_image(img)
        return embedding

    def _preprocess(self, img):
        img = to_pil(img)
        processed_img = self.tfms(img).unsqueeze(0).to(self.device)
        return processed_img

    def _configs(self):
        config = {}
        config['slip_vit_small'] = {}
        config['slip_vit_small']['weights'] = 'https://dl.fbaipublicfiles.com/slip/slip_small_100ep.pt' 
        config['slip_vit_base'] = {}
        config['slip_vit_base']['weights'] = 'https://dl.fbaipublicfiles.com/slip/slip_base_100ep.pt' 
        config['slip_vit_large'] = {}
        config['slip_vit_large']['weights'] = 'https://dl.fbaipublicfiles.com/slip/slip_large_100ep.pt'
        return config