wanghuaqiang commited on
Commit
4bbf816
1 Parent(s): 99eb82b

changed README.md

Browse files
Files changed (1) hide show
  1. README.md +47 -1
README.md CHANGED
@@ -1 +1,47 @@
1
- 基于有赞商品图片和标题语料训练的clip模型。
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!--
2
+ * @Description:
3
+ * @Version:
4
+ * @Author: Hardy
5
+ * @Date: 2022-02-09 15:13:53
6
+ * @LastEditors: Hardy
7
+ * @LastEditTime: 2022-02-09 16:59:01
8
+ -->
9
+
10
+ <br />
11
+ <p align="center">
12
+ <h1 align="center">clip-product-title-chinese</h1>
13
+ </p>
14
+
15
+ ## 基于有赞商品图片和标题语料训练的clip模型。
16
+
17
+
18
+ ## Usage
19
+ 使用模型前,请 git clone https://github.com/youzanai/trexpark.git
20
+
21
+
22
+ ```python
23
+ import torch
24
+ from src.clip.clip import ClipProcesserChinese, ClipChineseModel
25
+ import requests
26
+ from PIL import Image
27
+
28
+ clip_processor = ClipProcesserChinese.from_pretrained('youzanai/clip-product-title-chinese')
29
+ model = ClipChineseModel.from_pretrained('youzanai/clip-product-title-chinese')
30
+
31
+ url = 'http://img.yzcdn.cn/upload_files/2015/04/21/0140dac4657f874f2acff9294b28088c.jpg'
32
+ img = Image.open(requests.get(url, stream=True).raw).convert('RGB')
33
+ imgs = [img]
34
+ texts = ['运动鞋', '红色连衣裙', '黑色连衣裙', '大衣', '文具']
35
+
36
+ f = clip_processor(texts, imgs, return_tensors='pt', truncation=True, padding=True)
37
+ del f['token_type_ids']
38
+ with torch.no_grad():
39
+ out = model(**f)
40
+ logits_per_image, logits_per_text = out['logits_per_image'], out['logits_per_text']
41
+
42
+ print(logits_per_image.softmax(dim=-1).cpu().detach().numpy())
43
+
44
+ # 结果: [[1.1700666e-07 9.9948394e-01 5.1582896e-04 4.7687358e-11 6.9604440e-08]]
45
+ ```
46
+
47
+