blip
copied
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Readme
Files and versions
76 lines
2.9 KiB
76 lines
2.9 KiB
# 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
|
|
from towhee import register
|
|
from towhee.operator.base import NNOperator, OperatorFlag
|
|
from towhee.types.arg import arg, to_image_color
|
|
import torch
|
|
import ipdb
|
|
from towhee.types.image_utils import from_pil, to_pil
|
|
from torchvision import transforms
|
|
from torchvision.transforms.functional import InterpolationMode
|
|
|
|
@register(output_schema=['vec'])
|
|
class Blip(NNOperator):
|
|
"""
|
|
BLIP multi-modal embedding operator
|
|
"""
|
|
def __init__(self, model_name: str, modality: str):
|
|
super().__init__()
|
|
sys.path.append(str(Path(__file__).parent))
|
|
from models.blip import blip_feature_extractor
|
|
image_size = 224
|
|
model_url = 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_base.pth'
|
|
self.model = blip_feature_extractor(pretrained=model_url, image_size=image_size, vit='base')
|
|
|
|
self._modality = modality
|
|
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
|
self.tfms = transforms.Compose([
|
|
transforms.Resize((image_size,image_size),interpolation=InterpolationMode.BICUBIC),
|
|
transforms.ToTensor(),
|
|
transforms.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
|
|
])
|
|
|
|
def __call__(self, data):
|
|
ipdb.set_trace()
|
|
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):
|
|
text_feature = self.model(None, text, mode='text', device=self.device)[0,0]
|
|
return text_feature
|
|
|
|
@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)
|
|
img = self._preprocess(img)
|
|
caption = ''
|
|
image_feature = self.model(img, caption, mode='image', device=self.device)[0,0]
|
|
return image_feature
|
|
|
|
def _preprocess(self, img):
|
|
img = to_pil(img)
|
|
processed_img = self.tfms(img).unsqueeze(0).to(self.device)
|
|
return processed_img
|
|
|