Spaces:
Runtime error
Runtime error
Update off_topic.py
Browse files- off_topic.py +57 -10
off_topic.py
CHANGED
@@ -10,28 +10,71 @@ import numpy as np
|
|
10 |
import torch
|
11 |
import PIL
|
12 |
import imagehash
|
13 |
-
from transformers import CLIPModel, CLIPProcessor
|
14 |
from PIL import Image
|
15 |
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
class OffTopicDetector:
|
18 |
-
def __init__(self, model_id: str, device: Optional[str] = None, image_size: str = "E"):
|
19 |
self.device = device or ("cuda" if torch.cuda.is_available() else "cpu")
|
20 |
self.processor = CLIPProcessor.from_pretrained(model_id)
|
21 |
self.model = CLIPModel.from_pretrained(model_id).to(self.device)
|
22 |
self.image_size = image_size
|
|
|
23 |
|
24 |
def predict_probas(self, images: List[PIL.Image.Image], domain: str,
|
|
|
25 |
valid_templates: Optional[List[str]] = None,
|
26 |
invalid_classes: Optional[List[str]] = None,
|
27 |
autocast: bool = True):
|
|
|
|
|
28 |
if valid_templates:
|
29 |
valid_classes = [template.format(domain) for template in valid_templates]
|
30 |
else:
|
31 |
-
valid_classes = [f"a photo of {domain}", f"brochure with {domain} image", f"instructions for {domain}", f"{domain} diagram"
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
if not invalid_classes:
|
34 |
invalid_classes = ["promotional ad with store information", "promotional text", "google maps screenshot", "business card", "qr code"]
|
|
|
35 |
n_valid = len(valid_classes)
|
36 |
classes = valid_classes + invalid_classes
|
37 |
print(f"Valid classes: {valid_classes}", f"Invalid classes: {invalid_classes}", sep="\n")
|
@@ -59,18 +102,21 @@ class OffTopicDetector:
|
|
59 |
return probas, valid_probas, invalid_probas
|
60 |
|
61 |
def predict_probas_url(self, img_urls: List[str], domain: str,
|
|
|
62 |
valid_templates: Optional[List[str]] = None,
|
63 |
invalid_classes: Optional[List[str]] = None,
|
64 |
autocast: bool = True):
|
65 |
images = self.get_images(img_urls)
|
66 |
dedup_images = self._filter_dups(images)
|
67 |
-
return self.predict_probas(images, domain, valid_templates, invalid_classes, autocast)
|
68 |
|
69 |
def predict_probas_item(self, url_or_id: str,
|
|
|
70 |
valid_templates: Optional[List[str]] = None,
|
71 |
invalid_classes: Optional[List[str]] = None):
|
72 |
-
images, domain = self.get_item_data(url_or_id)
|
73 |
-
|
|
|
74 |
invalid_classes)
|
75 |
return images, domain, probas, valid_probas, invalid_probas
|
76 |
|
@@ -84,7 +130,8 @@ class OffTopicDetector:
|
|
84 |
item_id = re.sub("-", "", url_or_id)
|
85 |
start = time.time()
|
86 |
response = httpx.get(f"https://api.mercadolibre.com/items/{item_id}").json()
|
87 |
-
domain =
|
|
|
88 |
img_urls = [x["url"] for x in response["pictures"]]
|
89 |
img_urls = [x.replace("-O.jpg", f"-{self.image_size}.jpg") for x in img_urls]
|
90 |
end = time.time()
|
@@ -92,7 +139,7 @@ class OffTopicDetector:
|
|
92 |
print(f"Items API time: {round(duration * 1000, 0)} ms")
|
93 |
images = self.get_images(img_urls)
|
94 |
dedup_images = self._filter_dups(images)
|
95 |
-
return dedup_images, domain
|
96 |
|
97 |
def _filter_dups(self, images: List):
|
98 |
if len(images) > 1:
|
@@ -166,4 +213,4 @@ class OffTopicDetector:
|
|
166 |
if title:
|
167 |
fig.suptitle(title)
|
168 |
fig.tight_layout()
|
169 |
-
return
|
|
|
10 |
import torch
|
11 |
import PIL
|
12 |
import imagehash
|
13 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, CLIPModel, CLIPProcessor
|
14 |
from PIL import Image
|
15 |
|
16 |
|
17 |
+
class Translator:
|
18 |
+
def __init__(self, model_id: str, device: Optional[str] = None):
|
19 |
+
self.device = device or ("cuda" if torch.cuda.is_available() else "cpu")
|
20 |
+
self.model_id = model_id
|
21 |
+
self.tokenizer = AutoTokenizer.from_pretrained(
|
22 |
+
model_id)
|
23 |
+
self.model = AutoModelForSeq2SeqLM.from_pretrained(model_id).to(self.device)
|
24 |
+
self.bos_token_map = self.tokenizer.get_lang_id if hasattr(self.tokenizer, "get_lang_id") else self.tokenizer.lang_code_to_id
|
25 |
+
|
26 |
+
@property
|
27 |
+
def _language_code_mapper(self):
|
28 |
+
if "nllb" in self.model_id.lower():
|
29 |
+
return {"en": "eng_Latn",
|
30 |
+
"es": "spa_Latn",
|
31 |
+
"pt": "por_Latn"}
|
32 |
+
elif "m2m" in self.model_id.lower():
|
33 |
+
return {"en": "en",
|
34 |
+
"es": "es",
|
35 |
+
"pt": "pt"}
|
36 |
+
|
37 |
+
def translate(self, texts: List[str], src_lang: str, dest_lang: str = "en", max_length: int = 100):
|
38 |
+
self.tokenizer.src_lang = self._language_code_mapper[src_lang]
|
39 |
+
inputs = self.tokenizer(texts, return_tensors="pt").to(self.device)
|
40 |
+
translated_tokens = self.model.generate(
|
41 |
+
**inputs, forced_bos_token_id=self.bos_token_map["eng_Latn"], max_length=max_length
|
42 |
+
)
|
43 |
+
return self.tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)
|
44 |
+
|
45 |
+
|
46 |
class OffTopicDetector:
|
47 |
+
def __init__(self, model_id: str, device: Optional[str] = None, image_size: str = "E", translator: Optional[Translator] = None):
|
48 |
self.device = device or ("cuda" if torch.cuda.is_available() else "cpu")
|
49 |
self.processor = CLIPProcessor.from_pretrained(model_id)
|
50 |
self.model = CLIPModel.from_pretrained(model_id).to(self.device)
|
51 |
self.image_size = image_size
|
52 |
+
self.translator = translator
|
53 |
|
54 |
def predict_probas(self, images: List[PIL.Image.Image], domain: str,
|
55 |
+
title: Optional[str] = None,
|
56 |
valid_templates: Optional[List[str]] = None,
|
57 |
invalid_classes: Optional[List[str]] = None,
|
58 |
autocast: bool = True):
|
59 |
+
site, domain = domain.split("-")
|
60 |
+
domain = re.sub("_", " ", domain).lower()
|
61 |
if valid_templates:
|
62 |
valid_classes = [template.format(domain) for template in valid_templates]
|
63 |
else:
|
64 |
+
valid_classes = [f"a photo of {domain}", f"brochure with {domain} image", f"instructions for {domain}", f"{domain} diagram"]
|
65 |
+
if title:
|
66 |
+
if site == "CBT":
|
67 |
+
translated_title = title
|
68 |
+
else:
|
69 |
+
if site == "MLB":
|
70 |
+
src_lang = "pt"
|
71 |
+
else:
|
72 |
+
src_lang = "es"
|
73 |
+
translated_title = self.translator.translate(title, src_lang=src_lang, dest_lang="en", max_length=100)[0]
|
74 |
+
valid_classes.append(translated_title)
|
75 |
if not invalid_classes:
|
76 |
invalid_classes = ["promotional ad with store information", "promotional text", "google maps screenshot", "business card", "qr code"]
|
77 |
+
|
78 |
n_valid = len(valid_classes)
|
79 |
classes = valid_classes + invalid_classes
|
80 |
print(f"Valid classes: {valid_classes}", f"Invalid classes: {invalid_classes}", sep="\n")
|
|
|
102 |
return probas, valid_probas, invalid_probas
|
103 |
|
104 |
def predict_probas_url(self, img_urls: List[str], domain: str,
|
105 |
+
title: Optional[str] = None,
|
106 |
valid_templates: Optional[List[str]] = None,
|
107 |
invalid_classes: Optional[List[str]] = None,
|
108 |
autocast: bool = True):
|
109 |
images = self.get_images(img_urls)
|
110 |
dedup_images = self._filter_dups(images)
|
111 |
+
return self.predict_probas(images, domain, title, valid_templates, invalid_classes, autocast)
|
112 |
|
113 |
def predict_probas_item(self, url_or_id: str,
|
114 |
+
use_title: bool = False,
|
115 |
valid_templates: Optional[List[str]] = None,
|
116 |
invalid_classes: Optional[List[str]] = None):
|
117 |
+
images, domain, title = self.get_item_data(url_or_id)
|
118 |
+
title = title if use_title else None
|
119 |
+
probas, valid_probas, invalid_probas = self.predict_probas(images, domain, title, valid_templates,
|
120 |
invalid_classes)
|
121 |
return images, domain, probas, valid_probas, invalid_probas
|
122 |
|
|
|
130 |
item_id = re.sub("-", "", url_or_id)
|
131 |
start = time.time()
|
132 |
response = httpx.get(f"https://api.mercadolibre.com/items/{item_id}").json()
|
133 |
+
domain = response["domain_id"]
|
134 |
+
title = response["title"]
|
135 |
img_urls = [x["url"] for x in response["pictures"]]
|
136 |
img_urls = [x.replace("-O.jpg", f"-{self.image_size}.jpg") for x in img_urls]
|
137 |
end = time.time()
|
|
|
139 |
print(f"Items API time: {round(duration * 1000, 0)} ms")
|
140 |
images = self.get_images(img_urls)
|
141 |
dedup_images = self._filter_dups(images)
|
142 |
+
return dedup_images, domain, title
|
143 |
|
144 |
def _filter_dups(self, images: List):
|
145 |
if len(images) > 1:
|
|
|
213 |
if title:
|
214 |
fig.suptitle(title)
|
215 |
fig.tight_layout()
|
216 |
+
return
|