Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
classification code
Browse files- app.py +72 -52
- classifier.py +70 -0
app.py
CHANGED
@@ -31,6 +31,10 @@ S3_DATA_FOLDER = Path("sd-multiplayer-data")
|
|
31 |
|
32 |
DB_FOLDER = Path("diffusers-gallery-data")
|
33 |
|
|
|
|
|
|
|
|
|
34 |
s3 = boto3.client(service_name='s3',
|
35 |
aws_access_key_id=AWS_ACCESS_KEY_ID,
|
36 |
aws_secret_access_key=AWS_SECRET_KEY)
|
@@ -76,9 +80,9 @@ def fetch_models(page=0):
|
|
76 |
}
|
77 |
|
78 |
|
79 |
-
def fetch_model_card(
|
80 |
response = requests.get(
|
81 |
-
f'https://huggingface.co/{
|
82 |
return response.text
|
83 |
|
84 |
|
@@ -94,16 +98,31 @@ async def find_image_in_model_card(text):
|
|
94 |
return await asyncio.gather(*tasks)
|
95 |
|
96 |
|
97 |
-
def
|
98 |
-
|
99 |
-
|
100 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
|
102 |
-
|
103 |
-
|
|
|
|
|
|
|
104 |
|
|
|
|
|
|
|
|
|
105 |
|
106 |
-
|
|
|
107 |
initial = fetch_models(0)
|
108 |
num_pages = ceil(initial['numTotalItems'] / initial['numItemsPerPage'])
|
109 |
|
@@ -112,54 +131,55 @@ async def get_all_models():
|
|
112 |
print(f"Found {num_pages} pages")
|
113 |
|
114 |
# fetch all models
|
115 |
-
|
116 |
for page in tqdm(range(0, num_pages)):
|
117 |
print(f"Fetching page {page} of {num_pages}")
|
118 |
page_models = fetch_models(page)
|
119 |
-
|
120 |
-
|
121 |
-
with open(DB_FOLDER / "models_temp.json", "w") as f:
|
122 |
-
json.dump(models, f)
|
123 |
-
|
124 |
-
# fetch datacards and images
|
125 |
-
print(f"Found {len(models)} models")
|
126 |
-
final_models = []
|
127 |
-
for model in tqdm(models):
|
128 |
-
print(f"Fetching model {model['id']}")
|
129 |
-
model_card = fetch_model_card(model)
|
130 |
-
images = await find_image_in_model_card(model_card)
|
131 |
-
# style = await run_inference(f"https://api-inference.huggingface.co/models/{model['id']}", images[0])
|
132 |
-
style = []
|
133 |
-
# aesthetic = await run_inference(f"https://api-inference.huggingface.co/models/{model['id']}", images[0])
|
134 |
-
aesthetic = []
|
135 |
-
final_models.append(
|
136 |
-
{**model, "images": images, "style": style, "aesthetic": aesthetic}
|
137 |
-
)
|
138 |
-
return final_models
|
139 |
-
|
140 |
|
141 |
async def sync_data():
|
142 |
print("Fetching models")
|
143 |
-
|
144 |
-
|
|
|
145 |
with open(DB_FOLDER / "models.json", "w") as f:
|
146 |
-
json.dump(
|
147 |
# with open(DB_FOLDER / "models.json", "r") as f:
|
148 |
-
#
|
149 |
-
|
150 |
-
|
|
|
|
|
151 |
with database.get_db() as db:
|
152 |
cursor = db.cursor()
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
163 |
|
164 |
|
165 |
app = FastAPI()
|
@@ -174,7 +194,7 @@ app.add_middleware(
|
|
174 |
|
175 |
# @ app.get("/sync")
|
176 |
# async def sync(background_tasks: BackgroundTasks):
|
177 |
-
#
|
178 |
# return "Synced data to huggingface datasets"
|
179 |
|
180 |
|
@@ -189,16 +209,16 @@ def get_page(page: int = 1):
|
|
189 |
cursor.execute("""
|
190 |
SELECT *, COUNT(*) OVER() AS total
|
191 |
FROM models
|
192 |
-
WHERE json_extract(data, '$.likes') >
|
193 |
-
ORDER BY
|
194 |
LIMIT ? OFFSET ?
|
195 |
""", (MAX_PAGE_SIZE, (page - 1) * MAX_PAGE_SIZE))
|
196 |
results = cursor.fetchall()
|
197 |
-
total = results[0][
|
198 |
total_pages = (total + MAX_PAGE_SIZE - 1) // MAX_PAGE_SIZE
|
199 |
|
200 |
return {
|
201 |
-
"models": [json.loads(result[
|
202 |
"totalPages": total_pages
|
203 |
}
|
204 |
|
|
|
31 |
|
32 |
DB_FOLDER = Path("diffusers-gallery-data")
|
33 |
|
34 |
+
CLASSIFIER_URL = "https://radames-aesthetic-style-nsfw-classifier.hf.space/run/inference"
|
35 |
+
ASSETS_URL = "https://d26smi9133w0oo.cloudfront.net/diffusers-gallery/"
|
36 |
+
|
37 |
+
|
38 |
s3 = boto3.client(service_name='s3',
|
39 |
aws_access_key_id=AWS_ACCESS_KEY_ID,
|
40 |
aws_secret_access_key=AWS_SECRET_KEY)
|
|
|
80 |
}
|
81 |
|
82 |
|
83 |
+
def fetch_model_card(model_id):
|
84 |
response = requests.get(
|
85 |
+
f'https://huggingface.co/{model_id}/raw/main/README.md')
|
86 |
return response.text
|
87 |
|
88 |
|
|
|
98 |
return await asyncio.gather(*tasks)
|
99 |
|
100 |
|
101 |
+
def run_classifier(images):
|
102 |
+
images = [i for i in images if i is not None]
|
103 |
+
if len(images) > 0:
|
104 |
+
# classifying only the first image
|
105 |
+
images_urls = [ASSETS_URL + images[0]]
|
106 |
+
response = requests.post(CLASSIFIER_URL, json={"data": [
|
107 |
+
{"urls": images_urls}, # json urls: list of images urls
|
108 |
+
False, # enable/disable gallery image output
|
109 |
+
None, # single image input
|
110 |
+
None, # files input
|
111 |
+
]}).json()
|
112 |
|
113 |
+
# data response is array data:[[{img0}, {img1}, {img2}...], Label, Gallery],
|
114 |
+
class_data = response['data'][0][0]
|
115 |
+
print(class_data)
|
116 |
+
class_data_parsed = {row['label']: round(
|
117 |
+
row['score'], 3) for row in class_data}
|
118 |
|
119 |
+
# update row data with classificator data
|
120 |
+
return class_data_parsed
|
121 |
+
else:
|
122 |
+
return {}
|
123 |
|
124 |
+
|
125 |
+
async def get_all_new_models():
|
126 |
initial = fetch_models(0)
|
127 |
num_pages = ceil(initial['numTotalItems'] / initial['numItemsPerPage'])
|
128 |
|
|
|
131 |
print(f"Found {num_pages} pages")
|
132 |
|
133 |
# fetch all models
|
134 |
+
new_models = []
|
135 |
for page in tqdm(range(0, num_pages)):
|
136 |
print(f"Fetching page {page} of {num_pages}")
|
137 |
page_models = fetch_models(page)
|
138 |
+
new_models += page_models['models']
|
139 |
+
return new_models
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
140 |
|
141 |
async def sync_data():
|
142 |
print("Fetching models")
|
143 |
+
new_models = await get_all_new_models()
|
144 |
+
print(f"Found {len(new_models)} models")
|
145 |
+
# save list of all models for ids
|
146 |
with open(DB_FOLDER / "models.json", "w") as f:
|
147 |
+
json.dump(new_models, f)
|
148 |
# with open(DB_FOLDER / "models.json", "r") as f:
|
149 |
+
# new_models = json.load(f)
|
150 |
+
|
151 |
+
new_models_ids = [model['id'] for model in new_models]
|
152 |
+
|
153 |
+
# get existing models
|
154 |
with database.get_db() as db:
|
155 |
cursor = db.cursor()
|
156 |
+
cursor.execute("SELECT id FROM models")
|
157 |
+
existing_models = [row['id'] for row in cursor.fetchall()]
|
158 |
+
models_ids_to_add = list(set(new_models_ids) - set(existing_models))
|
159 |
+
# find all models id to add from new_models
|
160 |
+
models = [model for model in new_models if model['id'] in models_ids_to_add]
|
161 |
+
|
162 |
+
print(f"Found {len(models)} new models")
|
163 |
+
for model in tqdm(models):
|
164 |
+
model_id = model['id']
|
165 |
+
model_card = fetch_model_card(model_id)
|
166 |
+
images = await find_image_in_model_card(model_card)
|
167 |
+
classifier = run_classifier(images)
|
168 |
+
# update model row with image and classifier data
|
169 |
+
with database.get_db() as db:
|
170 |
+
cursor = db.cursor()
|
171 |
+
cursor.execute("INSERT INTO models(id, data) VALUES (?, ?)",
|
172 |
+
[model_id, json.dumps({
|
173 |
+
**model,
|
174 |
+
"images": images,
|
175 |
+
"class": classifier
|
176 |
+
})])
|
177 |
+
db.commit()
|
178 |
+
|
179 |
+
|
180 |
+
# print("Updating repository")
|
181 |
+
# subprocess.Popen(
|
182 |
+
# "git add . && git commit --amend -m 'update' && git push --force", cwd=DB_FOLDER, shell=True)
|
183 |
|
184 |
|
185 |
app = FastAPI()
|
|
|
194 |
|
195 |
# @ app.get("/sync")
|
196 |
# async def sync(background_tasks: BackgroundTasks):
|
197 |
+
# await sync_data()
|
198 |
# return "Synced data to huggingface datasets"
|
199 |
|
200 |
|
|
|
209 |
cursor.execute("""
|
210 |
SELECT *, COUNT(*) OVER() AS total
|
211 |
FROM models
|
212 |
+
WHERE json_extract(data, '$.likes') > 4
|
213 |
+
ORDER BY datetime(json_extract(data, '$.lastModified')) DESC
|
214 |
LIMIT ? OFFSET ?
|
215 |
""", (MAX_PAGE_SIZE, (page - 1) * MAX_PAGE_SIZE))
|
216 |
results = cursor.fetchall()
|
217 |
+
total = results[0]['total'] if results else 0
|
218 |
total_pages = (total + MAX_PAGE_SIZE - 1) // MAX_PAGE_SIZE
|
219 |
|
220 |
return {
|
221 |
+
"models": [json.loads(result['data']) for result in results],
|
222 |
"totalPages": total_pages
|
223 |
}
|
224 |
|
classifier.py
ADDED
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import re
|
3 |
+
import requests
|
4 |
+
import json
|
5 |
+
import subprocess
|
6 |
+
from io import BytesIO
|
7 |
+
import uuid
|
8 |
+
|
9 |
+
from math import ceil
|
10 |
+
from tqdm import tqdm
|
11 |
+
from pathlib import Path
|
12 |
+
|
13 |
+
from db import Database
|
14 |
+
|
15 |
+
DB_FOLDER = Path("diffusers-gallery-data")
|
16 |
+
|
17 |
+
database = Database(DB_FOLDER)
|
18 |
+
|
19 |
+
|
20 |
+
CLASSIFIER_URL = "https://radames-aesthetic-style-nsfw-classifier.hf.space/run/inference"
|
21 |
+
ASSETS_URL = "https://d26smi9133w0oo.cloudfront.net/diffusers-gallery/"
|
22 |
+
|
23 |
+
|
24 |
+
def main():
|
25 |
+
|
26 |
+
with database.get_db() as db:
|
27 |
+
cursor = db.cursor()
|
28 |
+
cursor.execute("""
|
29 |
+
SELECT *
|
30 |
+
FROM models
|
31 |
+
""")
|
32 |
+
results = list(cursor.fetchall())
|
33 |
+
|
34 |
+
for row in tqdm(results):
|
35 |
+
row_id = row['id']
|
36 |
+
# keep json data on row_data
|
37 |
+
row_data = json.loads(row['data'])
|
38 |
+
print("updating row", row_id)
|
39 |
+
images = row_data['images']
|
40 |
+
|
41 |
+
# filter nones
|
42 |
+
images = [i for i in images if i is not None]
|
43 |
+
if len(images) > 0:
|
44 |
+
# classifying only the first image
|
45 |
+
images_urls = [ASSETS_URL + images[0]]
|
46 |
+
response = requests.post(CLASSIFIER_URL, json={"data": [
|
47 |
+
{"urls": images_urls}, # json urls: list of images urls
|
48 |
+
False, # enable/disable gallery image output
|
49 |
+
None, # single image input
|
50 |
+
None, # files input
|
51 |
+
]}).json()
|
52 |
+
|
53 |
+
# data response is array data:[[{img0}, {img1}, {img2}...], Label, Gallery],
|
54 |
+
class_data = response['data'][0][0]
|
55 |
+
class_data_parsed = {row['label']: round(
|
56 |
+
row['score'], 3) for row in class_data}
|
57 |
+
|
58 |
+
# update row data with classificator data
|
59 |
+
row_data['class'] = class_data_parsed
|
60 |
+
else:
|
61 |
+
row_data['class'] = {}
|
62 |
+
with database.get_db() as db:
|
63 |
+
cursor = db.cursor()
|
64 |
+
cursor.execute("UPDATE models SET data = ? WHERE id = ?",
|
65 |
+
[json.dumps(row_data), row_id])
|
66 |
+
db.commit()
|
67 |
+
|
68 |
+
|
69 |
+
if __name__ == "__main__":
|
70 |
+
main()
|