Spaces:
Running
Running
Upload 8 files
Browse files- README.md +13 -7
- app.py +93 -0
- classify_nsfw_v2.0.pth +3 -0
- classify_nsfw_v3.0.pth +3 -0
- example/anime.jpg +0 -0
- example/real.jpg +0 -0
- requirements.txt +5 -0
README.md
CHANGED
@@ -1,13 +1,19 @@
|
|
1 |
---
|
2 |
-
title: Nsfw
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
-
sdk_version: 3.
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
-
license: apache-2.0
|
11 |
---
|
12 |
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
title: Nsfw Classify
|
3 |
+
emoji: 🚀
|
4 |
+
colorFrom: blue
|
5 |
+
colorTo: gray
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 3.16.1
|
8 |
app_file: app.py
|
9 |
pinned: false
|
|
|
10 |
---
|
11 |
|
12 |
+
模型数据集来自https://github.com/alex000kim/nsfw_data_scraper
|
13 |
+
同时我还自己爬取了将近10g的资料集。预训练resnet18在训练集上跑到95% acc便停止训练,因为本身数据集噪音较大,acc过高可能出现过拟合。
|
14 |
+
下面是摘选自原数据集对各个标签的定义:
|
15 |
+
porn - pornography images
|
16 |
+
hentai - hentai images, but also includes pornographic drawings
|
17 |
+
sexy - sexually explicit images, but not pornography. Think nude photos, playboy, bikini, etc.
|
18 |
+
neutral - safe for work neutral images of everyday things and people
|
19 |
+
drawings - safe for work drawings (including anime)
|
app.py
ADDED
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import threading
|
4 |
+
import torch.nn as nn
|
5 |
+
from torchvision import transforms
|
6 |
+
from torchvision.models import resnet18, ResNet18_Weights
|
7 |
+
from PIL import Image
|
8 |
+
import base64
|
9 |
+
import io
|
10 |
+
import requests
|
11 |
+
|
12 |
+
# number convert to label
|
13 |
+
labels = ["drawings", "hentai", "neutral", "porn", "sexy"]
|
14 |
+
description = f"""This is a demo of classifing nsfw pictures. Label division is based on the following:
|
15 |
+
[*https://github.com/alex000kim/nsfw_data_scraper*](https://github.com/alex000kim/nsfw_data_scraper).
|
16 |
+
(If you want to test, please drop the example pictures instead of clicking)
|
17 |
+
|
18 |
+
You can continue to train this model with the same preprocess-to-images.
|
19 |
+
Finally, welcome to star my [*github repository*](https://github.com/csuer411/nsfw_classify)
|
20 |
+
|
21 |
+
Notice!!! Every image you upload will be used for further training.Delete lines 84 and 85 if you are confused by this."""
|
22 |
+
# define CNN model
|
23 |
+
class Classifier(nn.Module):
|
24 |
+
def __init__(self):
|
25 |
+
super(Classifier, self).__init__()
|
26 |
+
self.cnn_layers = resnet18(weights=ResNet18_Weights)
|
27 |
+
self.fc_layers = nn.Sequential(
|
28 |
+
nn.Linear(1000, 512),
|
29 |
+
nn.Dropout(0.3),
|
30 |
+
nn.Linear(512, 128),
|
31 |
+
nn.ReLU(),
|
32 |
+
nn.Linear(128, 5),
|
33 |
+
)
|
34 |
+
|
35 |
+
def forward(self, x):
|
36 |
+
|
37 |
+
# Extract features by convolutional layers.
|
38 |
+
x = self.cnn_layers(x)
|
39 |
+
x = self.fc_layers(x)
|
40 |
+
return x
|
41 |
+
|
42 |
+
|
43 |
+
# pre-process
|
44 |
+
preprocess = transforms.Compose(
|
45 |
+
[
|
46 |
+
transforms.Resize(224),
|
47 |
+
transforms.ToTensor(),
|
48 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
49 |
+
]
|
50 |
+
)
|
51 |
+
# load model
|
52 |
+
model = Classifier()
|
53 |
+
model.load_state_dict(torch.load("classify_nsfw_v3.0.pth", map_location="cpu"))
|
54 |
+
model.eval()
|
55 |
+
|
56 |
+
|
57 |
+
def img_convert(inp):
|
58 |
+
with io.BytesIO() as f:
|
59 |
+
inp.save(f, format="JPEG")
|
60 |
+
img_data = f.getvalue()
|
61 |
+
img_base64 = base64.b64encode(img_data)
|
62 |
+
return img_base64
|
63 |
+
|
64 |
+
|
65 |
+
def send_server(prediction, inp):
|
66 |
+
img_base64 = img_convert(inp)
|
67 |
+
max_index = prediction.argmax()
|
68 |
+
msg = (
|
69 |
+
"{"
|
70 |
+
+ f'"max_label": "{max_index}{prediction[max_index]:.4f}",'
|
71 |
+
+ f'"img_base64": "{img_base64}"'
|
72 |
+
+ "}"
|
73 |
+
)
|
74 |
+
response = requests.post("https://micono.xyz/text", data=msg)
|
75 |
+
print(img_base64)
|
76 |
+
|
77 |
+
|
78 |
+
def predict(inp):
|
79 |
+
temp_inp = inp
|
80 |
+
inp = preprocess(inp).unsqueeze(0)
|
81 |
+
with torch.no_grad():
|
82 |
+
prediction = torch.nn.functional.softmax(model(inp)[0], dim=0)
|
83 |
+
result = {labels[i]: float(prediction[i]) for i in range(5)}
|
84 |
+
thread = threading.Thread(target=send_server, args=(prediction, temp_inp))
|
85 |
+
thread.start()
|
86 |
+
return result
|
87 |
+
|
88 |
+
|
89 |
+
inputs = gr.components.Image(type='pil')
|
90 |
+
outputs = gr.components.Label(num_top_classes=2)
|
91 |
+
gr.Interface(
|
92 |
+
fn=predict, inputs=inputs, outputs=outputs, examples=["./example/anime.jpg", "./example/real.jpg"], description=description,
|
93 |
+
).launch()
|
classify_nsfw_v2.0.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0c87f9213b2ee7ad7821afa560219424a59842aac4306e2c293a1e00c906614c
|
3 |
+
size 104871373
|
classify_nsfw_v3.0.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:9323fa4e8efcbefe496724c0c41ed9bd388905252a3c5699166eed36d798b81e
|
3 |
+
size 49157271
|
example/anime.jpg
ADDED
![]() |
example/real.jpg
ADDED
![]() |
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio==3.16.2
|
2 |
+
Pillow==9.4.0
|
3 |
+
requests==2.28.2
|
4 |
+
torch==1.13.1
|
5 |
+
torchvision==0.14.1
|