towhee
/
vit-image-embedding
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
64 lines
2.4 KiB
64 lines
2.4 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 timm
|
|
import torch
|
|
import numpy
|
|
from torch import nn as nn
|
|
from typing import NamedTuple
|
|
from towhee.operator import NNOperator
|
|
from timm.data import resolve_data_config
|
|
from timm.data.transforms_factory import create_transform
|
|
from towhee.utils.pil_utils import to_pil
|
|
|
|
import warnings
|
|
|
|
warnings.filterwarnings("ignore")
|
|
|
|
|
|
class VitImageEmbedding(NNOperator):
|
|
"""
|
|
Embedding extractor using ViT.
|
|
Args:
|
|
model_name (`string`):
|
|
Model name.
|
|
weights_path (`string`):
|
|
Path to local weights.
|
|
"""
|
|
|
|
def __init__(self, model_name: str = 'vit_large_patch16_224', num_classes: int = 1000,
|
|
framework: str = 'pytorch', weights_path: str = None) -> None:
|
|
super().__init__(framework=framework)
|
|
self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
|
if weights_path:
|
|
self.model = timm.create_model(model_name, checkpoint_path=weights_path, num_classes=num_classes)
|
|
else:
|
|
self.model = timm.create_model(model_name, pretrained=True, num_classes=num_classes)
|
|
self.model.eval()
|
|
config = resolve_data_config({}, model=self.model)
|
|
self.tfms = create_transform(**config)
|
|
|
|
def __call__(self, image: 'towhee.types.Image') -> NamedTuple('Outputs', [('feature_vector', numpy.ndarray)]):
|
|
img_tensor = self.tfms(to_pil(image)).unsqueeze(0)
|
|
Outputs = NamedTuple('Outputs', [('feature_vector', numpy.ndarray)])
|
|
self.model.to(self.device)
|
|
self.model.eval()
|
|
features = self.model.forward_features(img_tensor)
|
|
if features.dim() == 4: # if the shape of feature map is [N, C, H, W], where H > 1 and W > 1
|
|
global_pool = nn.AdaptiveAvgPool2d(1)
|
|
features = global_pool(features)
|
|
features = features.to('cpu')
|
|
features = features.flatten().detach().numpy()
|
|
return Outputs(features)
|