clip/clip.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.types.image_utils import to_pil
from towhee.operator.base import NNOperator, OperatorFlag
from towhee.types.arg import arg, to_image_color
from towhee import register
from towhee.models import clip


@register(output_schema=['vec'])
class Clip(NNOperator): 
    """
    CLIP multi-modal embedding operator
    """
    def __init__(self, model_name: str, modality: str):
        self.modality = modality
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.model = clip.create_model(model_name=model_name, pretrained=True, jit=True)
        self.tokenize = clip.tokenize
        self.tfms = transforms.Compose([
            transforms.Resize(224, interpolation=transforms.InterpolationMode.BICUBIC),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize(
               (0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
            ])

    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()

    def _inference_from_text(self, text):
        text = self.tokenize(text).to(self.device)
        text_features = self.model.encode_text(text)
        return text_features

    @arg(1, to_image_color('RGB'))
    def _inference_from_image(self, img):
        img = to_pil(img)
        image = self.tfms(img).unsqueeze(0).to(self.device)
        image_features = self.model.encode_image(image)
        return image_features