Spaces:
Sleeping
Sleeping
File size: 6,330 Bytes
f51b502 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 |
import os
from typing import Dict, List, Optional, Union
import pickle
import torch
import torchvision
from torch import nn
import tarfile
from PIL import Image
from torchvision import transforms
def read_im(input: Image.Image) -> Image.Image:
"""read im
Args:
input (Image.Image):
img
Returns:
Image.Image
"""
im = input
if not isinstance(im, Image.Image):
raise ValueError("""`input` should be a str or bytes or Image.Image!""")
im = im.convert("RGB")
return im
class Classifier(nn.Module):
"""Toxic Classifier.
Given a transformed image,`classifier` will get a toxic socre on it.
Attributes:
config (Optional[Dict],optional):
Modeling config.
Defaults to None.
"""
def __init__(self, config: Optional[Dict] = None) -> None:
super().__init__()
self.config = {} if config is None else config
self.resnet = torchvision.models.resnet50()
self.resnet.fc = nn.Linear(
in_features=self.config.get("in_features", 2048),
out_features=self.config.get("tag_num", 2),)
def forward(self, x: torch.Tensor) -> torch.Tensor:
out = self.resnet(x)
return out
@torch.no_grad()
def score(self, input: torch.Tensor) -> List[float]:
"""Scoring the input image(one input).
Args:
input (torch.Tensor):
img input(should be transformed).
Returns:
List[float]:
The toxic score of the input .
"""
return (
torch.softmax(self.forward(input), dim=1).detach().cpu().view(-1).tolist())
class Detector():
"""Toxic detector .
Attributes:
config (Optional[Dict],optional):
Modeling config.
Defaults to None.
"""
def __init__(self,*,config: Optional[Dict] = None,) -> None:
super().__init__()
if config is None:
config = {}
self._config = config
self._in_features = config.get("in_features", 2048)
self._tag_num = config.get("tag_num", 2)
self._tags = config.get("tags", ["obscene"])
self._classifier = Classifier(self.config)
self._trans = transforms.Compose(
[
# transforms.ToPILImage()
transforms.Resize(256),
transforms.CenterCrop(size=(224, 224)),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])
@property
def config(self):
return self._config
@config.setter
def config(self, config: Dict):
self._config = config
self._in_features = config.get("in_features", 2048)
self._tag_num = config.get("tag_num", 2)
self._tags = config.get("tags", ["obscene"])
@property
def classifier(self):
return self._classifier
def _load_pkl(self, path: str) -> Dict:
with open(path, "rb") as f:
file = pickle.load(f)
return file
def _unzip2dir(self, file: str, dir: Optional[str] = None) -> None:
if dir is None:
dir = self._tmpdir.name
if not os.path.isdir(dir):
raise ValueError("""`dir` shoud be a dir!""")
tar = tarfile.open(file, "r")
tar.extractall(path=dir)
tar.close()
def load(self, model: str) -> None:
"""Load state dict from local model path .
Args:
model (str):
Model file need to be loaded.
A string, the path of a pretrained model.
Raises:
ValueError: str model should be a path!
"""
if isinstance(model, str):
if os.path.isdir(model):
self._load_from_dir(model)
elif os.path.isfile(model):
dir = "./toxic_detection"
if os.path.exists(dir):
pass
else:
os.mkdir(dir)
self._unzip2dir(model, dir)
self._load_from_dir(dir)
else:
raise ValueError("""str model should be a path!""")
else:
raise ValueError("""str model should be a path!""")
def _load_from_dir(self, model_dir: str) -> None:
"""Set model params from `model_file`.
Args:
model_dir (str):
Dir containing model params.
"""
config = self._load_pkl(os.path.join(model_dir, "config.pkl"))
self.config = config
self._classifier = Classifier(config)
self._classifier.load_state_dict(
torch.load(os.path.join(model_dir, "classifier.pkl"), map_location="cpu"))
self._classifier.eval()
def _transform(self, input: Union[str, bytes, Image.Image]) -> torch.Tensor:
"""Transforms image to torch tensor.
Args:
input (Union[str,bytes,Image.Image]):
Image .
Raises:
ValueError:
`input` should be a str or bytes!
Returns:
torch.Tensor:
Transformed torch tensor.
"""
im = read_im(input)
out = self._trans(im).view(1, 3, 224, 224).float()
return out
def _score(self, input: torch.Tensor) -> List[float]:
"""Scoring the input image."""
toxic_score = self._classifier.score(input)
toxic_score = [round(s, 3) for s in toxic_score][1:]
return toxic_score
def detect(self, input: Union[str, bytes, Image.Image]) -> Dict:
"""Detects toxic contents from image `input`.
Args:
input (Union[str,bytes,Image.Image]):
Image path of bytes.
Raises:
ValueError:
`input` should be a str or bytes!
Returns:
Dict:
Pattern as {
"toxic_score " : Dict[str,float]
}.
"""
im = self._transform(input)
toxic_score = self._score(im)
out = {
"toxic_score": dict(
zip(
self._tags,
toxic_score,
)
),}
return out
|