使用 AutoMM 进行图像到图像语义匹配¶
计算两幅图像之间的相似度是计算机视觉中的常见任务，有多种实际应用，例如检测相同或不同的产品等。一般来说，图像相似度模型会将两幅图像作为输入并将其转换为向量，然后使用余弦相似度、点积或欧氏距离计算相似度得分来衡量两幅图像的相似或不同程度。
import os
import pandas as pd
import warnings
from IPython.display import Image, display
warnings.filterwarnings('ignore')
准备数据¶
在本教程中，我们将演示如何使用 AutoMM 对简化的 Stanford Online Products 数据集 (SOP) 进行图像到图像语义匹配。
Stanford Online Products 数据集被用于度量学习。此数据集中有 12 个产品类别：自行车、橱柜、椅子、咖啡机、风扇、水壶、台灯、马克杯、沙发、订书机、桌子和烤面包机。每个类别包含一些产品，每个产品都有从不同视角拍摄的几张图像。在这里，我们将同一产品的不同视角图像视为正样本对（标记为 1），将来自不同产品的图像视为负样本对（标记为 0）。
以下代码下载数据集并解压图像和标注文件。
download_dir = './ag_automm_tutorial_img2img'
zip_file = 'https://automl-mm-bench.s3.amazonaws.com/Stanford_Online_Products.zip'
from autogluon.core.utils.loaders import load_zip
load_zip.unzip(zip_file, unzip_dir=download_dir)
Downloading ./ag_automm_tutorial_img2img/file.zip from https://automl-mm-bench.s3.amazonaws.com/Stanford_Online_Products.zip...
0%| | 0.00/3.08G [00:00<?, ?iB/s]
0%| | 8.38M/3.08G [00:00<00:52, 58.6MiB/s]
0%| | 15.0M/3.08G [00:00<01:20, 38.0MiB/s]
1%| | 19.1M/3.08G [00:00<01:44, 29.3MiB/s]
1%| | 23.8M/3.08G [00:00<01:39, 30.7MiB/s]
1%| | 27.0M/3.08G [00:00<01:56, 26.3MiB/s]
1%| | 32.2M/3.08G [00:01<01:40, 30.4MiB/s]
1%| | 35.4M/3.08G [00:01<02:09, 23.5MiB/s]
1%|▏ | 41.9M/3.08G [00:01<01:41, 29.8MiB/s]
2%|▏ | 50.3M/3.08G [00:01<01:25, 35.3MiB/s]
2%|▏ | 57.0M/3.08G [00:01<01:32, 32.7MiB/s]
2%|▏ | 60.4M/3.08G [00:01<01:41, 29.7MiB/s]
2%|▏ | 67.1M/3.08G [00:02<01:31, 33.0MiB/s]
2%|▏ | 75.5M/3.08G [00:02<01:17, 39.0MiB/s]
3%|▎ | 82.5M/3.08G [00:02<01:37, 30.7MiB/s]
3%|▎ | 86.0M/3.08G [00:02<01:49, 27.5MiB/s]
3%|▎ | 91.3M/3.08G [00:02<01:33, 31.9MiB/s]
3%|▎ | 95.0M/3.08G [00:03<01:38, 30.3MiB/s]
3%|▎ | 101M/3.08G [00:03<01:36, 31.0MiB/s]
3%|▎ | 105M/3.08G [00:03<01:33, 31.7MiB/s]
4%|▎ | 109M/3.08G [00:03<01:31, 32.5MiB/s]
4%|▍ | 116M/3.08G [00:03<01:14, 39.7MiB/s]
4%|▍ | 120M/3.08G [00:03<01:24, 35.0MiB/s]
4%|▍ | 124M/3.08G [00:03<01:21, 36.3MiB/s]
4%|▍ | 128M/3.08G [00:03<01:29, 33.2MiB/s]
4%|▍ | 132M/3.08G [00:04<01:30, 32.5MiB/s]
4%|▍ | 136M/3.08G [00:04<01:43, 28.5MiB/s]
5%|▍ | 143M/3.08G [00:04<01:44, 28.2MiB/s]
5%|▍ | 149M/3.08G [00:04<01:25, 34.4MiB/s]
5%|▍ | 153M/3.08G [00:04<01:27, 33.7MiB/s]
5%|▌ | 159M/3.08G [00:04<01:18, 37.1MiB/s]
5%|▌ | 166M/3.08G [00:05<01:13, 39.5MiB/s]
6%|▌ | 170M/3.08G [00:05<01:20, 36.0MiB/s]
6%|▌ | 176M/3.08G [00:05<01:29, 32.4MiB/s]
6%|▌ | 185M/3.08G [00:05<01:14, 38.8MiB/s]
6%|▌ | 191M/3.08G [00:05<01:06, 43.4MiB/s]
6%|▋ | 196M/3.08G [00:05<01:14, 38.8MiB/s]
7%|▋ | 201M/3.08G [00:05<01:09, 41.7MiB/s]
7%|▋ | 208M/3.08G [00:06<01:09, 41.5MiB/s]
7%|▋ | 212M/3.08G [00:06<01:48, 26.6MiB/s]
7%|▋ | 218M/3.08G [00:06<01:38, 29.2MiB/s]
7%|▋ | 225M/3.08G [00:06<01:22, 34.8MiB/s]
7%|▋ | 229M/3.08G [00:06<01:33, 30.4MiB/s]
8%|▊ | 233M/3.08G [00:07<01:35, 29.7MiB/s]
8%|▊ | 236M/3.08G [00:07<01:48, 26.3MiB/s]
8%|▊ | 242M/3.08G [00:07<02:02, 23.1MiB/s]
8%|▊ | 244M/3.08G [00:07<02:00, 23.5MiB/s]
8%|▊ | 252M/3.08G [00:07<01:36, 29.3MiB/s]
8%|▊ | 259M/3.08G [00:07<01:14, 38.0MiB/s]
9%|▊ | 263M/3.08G [00:08<01:26, 32.6MiB/s]
9%|▊ | 268M/3.08G [00:08<01:39, 28.2MiB/s]
9%|▉ | 275M/3.08G [00:08<01:25, 33.0MiB/s]
9%|▉ | 279M/3.08G [00:08<01:28, 31.8MiB/s]
9%|▉ | 285M/3.08G [00:08<01:20, 34.6MiB/s]
9%|▉ | 292M/3.08G [00:08<01:05, 42.5MiB/s]
10%|▉ | 297M/3.08G [00:09<01:11, 39.1MiB/s]
10%|▉ | 302M/3.08G [00:09<01:18, 35.2MiB/s]
10%|█ | 310M/3.08G [00:09<01:11, 38.6MiB/s]
10%|█ | 314M/3.08G [00:09<01:19, 34.7MiB/s]
10%|█ | 319M/3.08G [00:09<01:31, 30.2MiB/s]
11%|█ | 325M/3.08G [00:09<01:24, 32.7MiB/s]
11%|█ | 329M/3.08G [00:10<01:30, 30.5MiB/s]
11%|█ | 336M/3.08G [00:10<01:17, 35.6MiB/s]
11%|█ | 342M/3.08G [00:10<01:22, 33.1MiB/s]
11%|█ | 346M/3.08G [00:10<01:26, 31.6MiB/s]
11%|█▏ | 352M/3.08G [00:10<01:14, 36.4MiB/s]
12%|█▏ | 359M/3.08G [00:10<01:16, 35.8MiB/s]
12%|█▏ | 363M/3.08G [00:11<01:24, 32.2MiB/s]
12%|█▏ | 368M/3.08G [00:11<01:35, 28.6MiB/s]
12%|█▏ | 371M/3.08G [00:11<01:49, 24.9MiB/s]
12%|█▏ | 377M/3.08G [00:11<01:40, 26.8MiB/s]
13%|█▎ | 386M/3.08G [00:11<01:21, 33.0MiB/s]
13%|█▎ | 394M/3.08G [00:11<01:03, 42.1MiB/s]
13%|█▎ | 399M/3.08G [00:12<01:13, 36.7MiB/s]
13%|█▎ | 403M/3.08G [00:12<01:17, 34.8MiB/s]
13%|█▎ | 411M/3.08G [00:12<01:02, 42.7MiB/s]
14%|█▎ | 418M/3.08G [00:12<01:29, 29.8MiB/s]
14%|█▎ | 421M/3.08G [00:12<01:29, 29.7MiB/s]
14%|█▍ | 426M/3.08G [00:13<01:25, 31.0MiB/s]
14%|█▍ | 430M/3.08G [00:13<01:29, 29.6MiB/s]
14%|█▍ | 434M/3.08G [00:13<01:20, 32.7MiB/s]
14%|█▍ | 438M/3.08G [00:13<01:22, 32.1MiB/s]
14%|█▍ | 445M/3.08G [00:13<01:08, 38.5MiB/s]
15%|█▍ | 449M/3.08G [00:13<01:17, 34.0MiB/s]
15%|█▍ | 453M/3.08G [00:13<01:25, 30.8MiB/s]
15%|█▍ | 461M/3.08G [00:14<01:14, 35.4MiB/s]
15%|█▌ | 468M/3.08G [00:14<01:06, 39.4MiB/s]
15%|█▌ | 472M/3.08G [00:14<01:11, 36.4MiB/s]
16%|█▌ | 478M/3.08G [00:14<01:13, 35.6MiB/s]
16%|█▌ | 485M/3.08G [00:14<01:31, 28.3MiB/s]
16%|█▌ | 488M/3.08G [00:14<01:33, 27.7MiB/s]
16%|█▌ | 495M/3.08G [00:15<01:22, 31.5MiB/s]
16%|█▋ | 503M/3.08G [00:15<01:11, 36.0MiB/s]
17%|█▋ | 509M/3.08G [00:15<01:03, 40.7MiB/s]
17%|█▋ | 514M/3.08G [00:15<01:12, 35.4MiB/s]
17%|█▋ | 518M/3.08G [00:15<01:08, 37.5MiB/s]
17%|█▋ | 522M/3.08G [00:15<01:22, 31.1MiB/s]
17%|█▋ | 528M/3.08G [00:16<02:03, 20.6MiB/s]
17%|█▋ | 535M/3.08G [00:16<02:28, 17.1MiB/s]
17%|█▋ | 538M/3.08G [00:17<02:36, 16.3MiB/s]
18%|█▊ | 544M/3.08G [00:17<02:01, 20.8MiB/s]
18%|█▊ | 547M/3.08G [00:17<01:52, 22.6MiB/s]
18%|█▊ | 552M/3.08G [00:17<01:39, 25.6MiB/s]
18%|█▊ | 555M/3.08G [00:17<01:44, 24.1MiB/s]
18%|█▊ | 562M/3.08G [00:17<01:26, 29.2MiB/s]
18%|█▊ | 569M/3.08G [00:17<01:22, 30.4MiB/s]
19%|█▊ | 572M/3.08G [00:18<01:31, 27.5MiB/s]
19%|█▊ | 577M/3.08G [00:18<01:17, 32.2MiB/s]
19%|█▉ | 581M/3.08G [00:18<01:28, 28.2MiB/s]
19%|█▉ | 586M/3.08G [00:18<01:21, 30.8MiB/s]
19%|█▉ | 589M/3.08G [00:18<01:26, 28.9MiB/s]
19%|█▉ | 594M/3.08G [00:18<01:34, 26.3MiB/s]
19%|█▉ | 597M/3.08G [00:19<01:44, 23.8MiB/s]
20%|█▉ | 602M/3.08G [00:19<01:42, 24.1MiB/s]
20%|█▉ | 605M/3.08G [00:19<02:10, 19.0MiB/s]
20%|█▉ | 611M/3.08G [00:19<01:33, 26.3MiB/s]
20%|█▉ | 614M/3.08G [00:19<01:33, 26.4MiB/s]
20%|██ | 619M/3.08G [00:19<01:33, 26.3MiB/s]
20%|██ | 622M/3.08G [00:20<01:51, 22.0MiB/s]
20%|██ | 627M/3.08G [00:20<01:29, 27.4MiB/s]
20%|██ | 631M/3.08G [00:20<01:36, 25.4MiB/s]
21%|██ | 638M/3.08G [00:20<01:16, 32.0MiB/s]
21%|██ | 645M/3.08G [00:20<01:14, 32.9MiB/s]
21%|██ | 648M/3.08G [00:20<01:18, 30.9MiB/s]
21%|██ | 652M/3.08G [00:20<01:14, 32.4MiB/s]
21%|██ | 655M/3.08G [00:21<01:20, 30.1MiB/s]
21%|██▏ | 663M/3.08G [00:21<01:21, 29.9MiB/s]
22%|██▏ | 669M/3.08G [00:21<01:05, 37.0MiB/s]
22%|██▏ | 673M/3.08G [00:21<01:14, 32.2MiB/s]
22%|██▏ | 679M/3.08G [00:21<01:04, 37.1MiB/s]
22%|██▏ | 688M/3.08G [00:21<00:59, 40.1MiB/s]
23%|██▎ | 696M/3.08G [00:22<00:49, 48.5MiB/s]
23%|██▎ | 703M/3.08G [00:22<01:02, 37.9MiB/s]
23%|██▎ | 707M/3.08G [00:22<01:08, 34.7MiB/s]
23%|██▎ | 713M/3.08G [00:22<01:12, 32.9MiB/s]
23%|██▎ | 720M/3.08G [00:22<01:01, 38.4MiB/s]
23%|██▎ | 724M/3.08G [00:23<01:22, 28.8MiB/s]
24%|██▎ | 728M/3.08G [00:23<01:17, 30.5MiB/s]
24%|██▎ | 732M/3.08G [00:23<01:18, 30.2MiB/s]
24%|██▍ | 738M/3.08G [00:23<01:06, 35.5MiB/s]
24%|██▍ | 745M/3.08G [00:23<00:55, 41.9MiB/s]
24%|██▍ | 749M/3.08G [00:23<00:55, 42.4MiB/s]
24%|██▍ | 755M/3.08G [00:23<00:53, 43.6MiB/s]
25%|██▍ | 762M/3.08G [00:23<00:58, 39.7MiB/s]
25%|██▍ | 766M/3.08G [00:24<01:22, 28.0MiB/s]
25%|██▍ | 770M/3.08G [00:24<01:22, 28.0MiB/s]
25%|██▌ | 773M/3.08G [00:24<01:33, 24.7MiB/s]
25%|██▌ | 780M/3.08G [00:24<01:25, 27.1MiB/s]
26%|██▌ | 789M/3.08G [00:25<01:12, 31.8MiB/s]
26%|██▌ | 797M/3.08G [00:25<01:05, 34.7MiB/s]
26%|██▌ | 804M/3.08G [00:25<00:59, 38.2MiB/s]
26%|██▌ | 808M/3.08G [00:25<01:05, 34.9MiB/s]
26%|██▋ | 814M/3.08G [00:25<00:56, 40.3MiB/s]
27%|██▋ | 822M/3.08G [00:25<00:45, 49.8MiB/s]
27%|██▋ | 827M/3.08G [00:25<00:51, 44.2MiB/s]
27%|██▋ | 832M/3.08G [00:26<00:56, 39.7MiB/s]
27%|██▋ | 839M/3.08G [00:26<00:52, 43.0MiB/s]
27%|██▋ | 846M/3.08G [00:26<00:46, 48.4MiB/s]
28%|██▊ | 851M/3.08G [00:26<00:52, 42.3MiB/s]
28%|██▊ | 856M/3.08G [00:26<01:02, 35.7MiB/s]
28%|██▊ | 862M/3.08G [00:26<01:07, 32.7MiB/s]
28%|██▊ | 866M/3.08G [00:27<01:13, 30.4MiB/s]
28%|██▊ | 872M/3.08G [00:27<01:13, 30.1MiB/s]
29%|██▊ | 881M/3.08G [00:27<00:55, 39.6MiB/s]
29%|██▊ | 885M/3.08G [00:27<00:54, 40.4MiB/s]
29%|██▉ | 890M/3.08G [00:27<01:01, 35.5MiB/s]
29%|██▉ | 898M/3.08G [00:27<00:52, 41.4MiB/s]
29%|██▉ | 906M/3.08G [00:27<00:50, 42.9MiB/s]
30%|██▉ | 913M/3.08G [00:28<00:45, 47.6MiB/s]
30%|██▉ | 918M/3.08G [00:28<00:47, 45.6MiB/s]
30%|██▉ | 922M/3.08G [00:28<01:06, 32.6MiB/s]
30%|███ | 926M/3.08G [00:28<01:07, 32.0MiB/s]
30%|███ | 931M/3.08G [00:28<01:03, 34.1MiB/s]
30%|███ | 934M/3.08G [00:28<01:08, 31.2MiB/s]
30%|███ | 940M/3.08G [00:29<01:20, 26.7MiB/s]
31%|███ | 946M/3.08G [00:29<01:09, 30.8MiB/s]
31%|███ | 949M/3.08G [00:29<01:27, 24.3MiB/s]
31%|███ | 955M/3.08G [00:29<01:35, 22.4MiB/s]
31%|███ | 957M/3.08G [00:29<01:57, 18.2MiB/s]
31%|███▏ | 965M/3.08G [00:30<01:31, 23.2MiB/s]
31%|███▏ | 971M/3.08G [00:30<01:20, 26.2MiB/s]
32%|███▏ | 974M/3.08G [00:30<01:31, 23.0MiB/s]
32%|███▏ | 981M/3.08G [00:30<01:06, 31.5MiB/s]
32%|███▏ | 985M/3.08G [00:30<01:12, 29.0MiB/s]
32%|███▏ | 989M/3.08G [00:31<01:25, 24.6MiB/s]
32%|███▏ | 991M/3.08G [00:31<01:30, 23.1MiB/s]
32%|███▏ | 994M/3.08G [00:31<01:46, 19.6MiB/s]
32%|███▏ | 996M/3.08G [00:31<01:43, 20.1MiB/s]
32%|███▏ | 999M/3.08G [00:31<01:49, 19.0MiB/s]
33%|███▎ | 1.01G/3.08G [00:31<01:21, 25.5MiB/s]
33%|███▎ | 1.02G/3.08G [00:32<01:02, 33.1MiB/s]
33%|███▎ | 1.02G/3.08G [00:32<01:02, 32.9MiB/s]
33%|███▎ | 1.02G/3.08G [00:32<01:11, 28.6MiB/s]
33%|███▎ | 1.03G/3.08G [00:32<01:07, 30.3MiB/s]
34%|███▎ | 1.04G/3.08G [00:32<00:57, 35.5MiB/s]
34%|███▍ | 1.04G/3.08G [00:32<01:02, 32.8MiB/s]
34%|███▍ | 1.05G/3.08G [00:33<01:02, 32.4MiB/s]
34%|███▍ | 1.06G/3.08G [00:33<00:57, 35.4MiB/s]
34%|███▍ | 1.06G/3.08G [00:33<00:51, 39.0MiB/s]
35%|███▍ | 1.07G/3.08G [00:33<00:52, 38.7MiB/s]
35%|███▍ | 1.07G/3.08G [00:33<00:59, 33.9MiB/s]
35%|███▍ | 1.08G/3.08G [00:33<01:04, 31.0MiB/s]
35%|███▌ | 1.08G/3.08G [00:33<01:00, 33.2MiB/s]
35%|███▌ | 1.08G/3.08G [00:34<01:06, 30.3MiB/s]
35%|███▌ | 1.09G/3.08G [00:34<01:05, 30.4MiB/s]
35%|███▌ | 1.09G/3.08G [00:34<01:29, 22.3MiB/s]
36%|███▌ | 1.10G/3.08G [00:34<01:09, 28.5MiB/s]
36%|███▌ | 1.11G/3.08G [00:34<00:50, 39.2MiB/s]
36%|███▌ | 1.11G/3.08G [00:35<00:57, 34.4MiB/s]
36%|███▌ | 1.12G/3.08G [00:35<01:15, 26.0MiB/s]
36%|███▋ | 1.12G/3.08G [00:35<01:05, 29.9MiB/s]
37%|███▋ | 1.13G/3.08G [00:35<00:52, 37.4MiB/s]
37%|███▋ | 1.14G/3.08G [00:35<01:05, 29.6MiB/s]
37%|███▋ | 1.14G/3.08G [00:35<00:59, 32.9MiB/s]
37%|███▋ | 1.14G/3.08G [00:36<01:10, 27.7MiB/s]
37%|███▋ | 1.15G/3.08G [00:36<01:12, 26.8MiB/s]
37%|███▋ | 1.16G/3.08G [00:36<01:02, 30.9MiB/s]
38%|███▊ | 1.16G/3.08G [00:36<01:07, 28.5MiB/s]
38%|███▊ | 1.16G/3.08G [00:36<01:21, 23.7MiB/s]
38%|███▊ | 1.17G/3.08G [00:37<01:25, 22.3MiB/s]
38%|███▊ | 1.17G/3.08G [00:37<01:18, 24.4MiB/s]
38%|███▊ | 1.18G/3.08G [00:37<01:27, 21.7MiB/s]
38%|███▊ | 1.18G/3.08G [00:37<01:09, 27.5MiB/s]
39%|███▊ | 1.19G/3.08G [00:37<00:57, 32.7MiB/s]
39%|███▉ | 1.20G/3.08G [00:37<00:49, 38.3MiB/s]
39%|███▉ | 1.20G/3.08G [00:38<00:49, 37.7MiB/s]
39%|███▉ | 1.21G/3.08G [00:38<00:46, 40.5MiB/s]
39%|███▉ | 1.21G/3.08G [00:38<00:59, 31.4MiB/s]
39%|███▉ | 1.22G/3.08G [00:38<01:03, 29.2MiB/s]
40%|███▉ | 1.22G/3.08G [00:38<00:53, 35.0MiB/s]
40%|███▉ | 1.23G/3.08G [00:38<00:46, 39.9MiB/s]
40%|████ | 1.24G/3.08G [00:39<00:42, 43.1MiB/s]
40%|████ | 1.25G/3.08G [00:39<00:48, 37.8MiB/s]
41%|████ | 1.25G/3.08G [00:39<00:55, 33.0MiB/s]
41%|████ | 1.26G/3.08G [00:39<00:53, 34.1MiB/s]
41%|████ | 1.27G/3.08G [00:39<00:44, 40.9MiB/s]
41%|████ | 1.27G/3.08G [00:39<00:51, 35.2MiB/s]
41%|████▏ | 1.27G/3.08G [00:40<00:52, 34.6MiB/s]
41%|████▏ | 1.28G/3.08G [00:40<00:53, 34.1MiB/s]
42%|████▏ | 1.28G/3.08G [00:40<00:51, 34.7MiB/s]
42%|████▏ | 1.29G/3.08G [00:40<00:50, 35.7MiB/s]
42%|████▏ | 1.30G/3.08G [00:40<00:48, 36.8MiB/s]
42%|████▏ | 1.31G/3.08G [00:40<00:42, 41.9MiB/s]
43%|████▎ | 1.31G/3.08G [00:41<00:48, 36.9MiB/s]
43%|████▎ | 1.32G/3.08G [00:41<00:52, 33.4MiB/s]
43%|████▎ | 1.33G/3.08G [00:41<00:47, 36.9MiB/s]
43%|████▎ | 1.33G/3.08G [00:41<00:46, 37.3MiB/s]
43%|████▎ | 1.34G/3.08G [00:41<00:52, 33.5MiB/s]
44%|████▎ | 1.34G/3.08G [00:42<00:58, 29.7MiB/s]
44%|████▍ | 1.35G/3.08G [00:42<00:53, 32.1MiB/s]
44%|████▍ | 1.36G/3.08G [00:42<00:49, 35.1MiB/s]
44%|████▍ | 1.36G/3.08G [00:42<00:54, 31.8MiB/s]
44%|████▍ | 1.37G/3.08G [00:42<00:44, 38.3MiB/s]
44%|████▍ | 1.37G/3.08G [00:42<00:54, 31.4MiB/s]
45%|████▍ | 1.38G/3.08G [00:43<00:51, 33.4MiB/s]
45%|████▍ | 1.38G/3.08G [00:43<00:44, 37.8MiB/s]
45%|████▌ | 1.39G/3.08G [00:43<00:48, 34.7MiB/s]
45%|████▌ | 1.39G/3.08G [00:43<00:51, 32.7MiB/s]
45%|████▌ | 1.40G/3.08G [00:43<00:56, 30.0MiB/s]
45%|████▌ | 1.40G/3.08G [00:43<01:08, 24.5MiB/s]
46%|████▌ | 1.41G/3.08G [00:44<00:56, 29.4MiB/s]
46%|████▌ | 1.41G/3.08G [00:44<01:00, 27.8MiB/s]
46%|████▌ | 1.42G/3.08G [00:44<00:50, 32.8MiB/s]
46%|████▌ | 1.43G/3.08G [00:44<00:45, 36.1MiB/s]
46%|████▋ | 1.43G/3.08G [00:44<00:43, 38.2MiB/s]
47%|████▋ | 1.44G/3.08G [00:44<00:47, 34.9MiB/s]
47%|████▋ | 1.44G/3.08G [00:45<00:46, 35.3MiB/s]
47%|████▋ | 1.45G/3.08G [00:45<00:44, 37.0MiB/s]
47%|████▋ | 1.45G/3.08G [00:45<00:53, 30.8MiB/s]
47%|████▋ | 1.46G/3.08G [00:45<00:57, 28.1MiB/s]
47%|████▋ | 1.46G/3.08G [00:45<00:58, 27.6MiB/s]
48%|████▊ | 1.47G/3.08G [00:45<00:54, 29.9MiB/s]
48%|████▊ | 1.47G/3.08G [00:46<01:01, 26.2MiB/s]
48%|████▊ | 1.48G/3.08G [00:46<00:46, 34.6MiB/s]
48%|████▊ | 1.48G/3.08G [00:46<00:52, 30.4MiB/s]
48%|████▊ | 1.48G/3.08G [00:46<00:52, 30.7MiB/s]
48%|████▊ | 1.49G/3.08G [00:46<00:54, 29.4MiB/s]
48%|████▊ | 1.49G/3.08G [00:46<00:53, 29.6MiB/s]
48%|████▊ | 1.49G/3.08G [00:46<01:08, 23.2MiB/s]
49%|████▊ | 1.50G/3.08G [00:47<00:45, 35.2MiB/s]
49%|████▉ | 1.51G/3.08G [00:47<00:56, 28.2MiB/s]
49%|████▉ | 1.51G/3.08G [00:47<01:01, 25.6MiB/s]
49%|████▉ | 1.52G/3.08G [00:47<00:49, 31.6MiB/s]
49%|████▉ | 1.52G/3.08G [00:47<00:58, 26.6MiB/s]
49%|████▉ | 1.52G/3.08G [00:47<00:52, 29.4MiB/s]
50%|████▉ | 1.53G/3.08G [00:48<00:54, 28.7MiB/s]
50%|████▉ | 1.53G/3.08G [00:48<00:46, 33.4MiB/s]
50%|████▉ | 1.54G/3.08G [00:48<00:51, 30.0MiB/s]
50%|█████ | 1.54G/3.08G [00:48<00:49, 31.4MiB/s]
50%|█████ | 1.55G/3.08G [00:48<00:43, 35.4MiB/s]
50%|█████ | 1.55G/3.08G [00:48<00:49, 30.8MiB/s]
51%|█████ | 1.56G/3.08G [00:48<00:44, 34.4MiB/s]
51%|█████ | 1.57G/3.08G [00:49<00:37, 40.7MiB/s]
51%|█████ | 1.58G/3.08G [00:49<00:32, 46.6MiB/s]
51%|█████▏ | 1.58G/3.08G [00:49<00:34, 43.3MiB/s]
51%|█████▏ | 1.59G/3.08G [00:49<00:36, 40.5MiB/s]
52%|█████▏ | 1.59G/3.08G [00:49<00:35, 41.7MiB/s]
52%|█████▏ | 1.60G/3.08G [00:49<00:43, 34.0MiB/s]
52%|█████▏ | 1.60G/3.08G [00:49<00:40, 36.2MiB/s]
52%|█████▏ | 1.60G/3.08G [00:50<00:46, 31.9MiB/s]
52%|█████▏ | 1.61G/3.08G [00:50<00:55, 26.6MiB/s]
52%|█████▏ | 1.61G/3.08G [00:50<00:56, 26.2MiB/s]
52%|█████▏ | 1.62G/3.08G [00:50<00:47, 31.0MiB/s]
53%|█████▎ | 1.63G/3.08G [00:50<00:43, 33.8MiB/s]
53%|█████▎ | 1.63G/3.08G [00:50<00:36, 39.5MiB/s]
53%|█████▎ | 1.64G/3.08G [00:51<00:38, 37.9MiB/s]
53%|█████▎ | 1.64G/3.08G [00:51<00:34, 41.7MiB/s]
54%|█████▎ | 1.65G/3.08G [00:51<00:33, 42.4MiB/s]
54%|█████▍ | 1.66G/3.08G [00:51<00:33, 41.9MiB/s]
54%|█████▍ | 1.67G/3.08G [00:51<00:31, 44.9MiB/s]
54%|█████▍ | 1.67G/3.08G [00:51<00:32, 43.6MiB/s]
54%|█████▍ | 1.68G/3.08G [00:51<00:33, 41.6MiB/s]
55%|█████▍ | 1.69G/3.08G [00:52<00:32, 43.3MiB/s]
55%|█████▍ | 1.69G/3.08G [00:52<00:31, 43.8MiB/s]
55%|█████▌ | 1.70G/3.08G [00:52<00:35, 38.7MiB/s]
55%|█████▌ | 1.70G/3.08G [00:52<00:36, 38.3MiB/s]
55%|█████▌ | 1.71G/3.08G [00:52<00:31, 43.7MiB/s]
56%|█████▌ | 1.71G/3.08G [00:52<00:41, 33.3MiB/s]
56%|█████▌ | 1.72G/3.08G [00:53<00:43, 31.6MiB/s]
56%|█████▌ | 1.72G/3.08G [00:53<00:48, 27.9MiB/s]
56%|█████▌ | 1.73G/3.08G [00:53<00:42, 31.8MiB/s]
56%|█████▌ | 1.73G/3.08G [00:53<00:51, 26.1MiB/s]
56%|█████▋ | 1.74G/3.08G [00:53<00:46, 29.1MiB/s]
57%|█████▋ | 1.74G/3.08G [00:53<00:35, 37.6MiB/s]
57%|█████▋ | 1.75G/3.08G [00:54<00:35, 37.4MiB/s]
57%|█████▋ | 1.75G/3.08G [00:54<00:37, 35.2MiB/s]
57%|█████▋ | 1.76G/3.08G [00:54<00:30, 42.9MiB/s]
57%|█████▋ | 1.76G/3.08G [00:54<00:32, 40.5MiB/s]
57%|█████▋ | 1.77G/3.08G [00:54<00:39, 33.0MiB/s]
57%|█████▋ | 1.77G/3.08G [00:54<00:45, 29.1MiB/s]
58%|█████▊ | 1.78G/3.08G [00:54<00:44, 29.6MiB/s]
58%|█████▊ | 1.79G/3.08G [00:55<00:35, 36.3MiB/s]
58%|█████▊ | 1.79G/3.08G [00:55<00:42, 30.2MiB/s]
58%|█████▊ | 1.80G/3.08G [00:55<00:46, 27.9MiB/s]
58%|█████▊ | 1.80G/3.08G [00:55<00:42, 29.9MiB/s]
59%|█████▊ | 1.80G/3.08G [00:55<00:43, 29.1MiB/s]
59%|█████▉ | 1.81G/3.08G [00:56<00:42, 29.9MiB/s]
59%|█████▉ | 1.82G/3.08G [00:56<00:39, 31.6MiB/s]
59%|█████▉ | 1.83G/3.08G [00:56<00:36, 34.5MiB/s]
60%|█████▉ | 1.84G/3.08G [00:56<00:34, 35.7MiB/s]
60%|█████▉ | 1.84G/3.08G [00:56<00:33, 37.0MiB/s]
60%|█████▉ | 1.85G/3.08G [00:57<00:34, 36.1MiB/s]
60%|██████ | 1.85G/3.08G [00:57<00:35, 34.2MiB/s]
60%|██████ | 1.86G/3.08G [00:57<00:31, 38.3MiB/s]
60%|██████ | 1.86G/3.08G [00:57<00:32, 37.0MiB/s]
61%|██████ | 1.87G/3.08G [00:57<00:35, 34.2MiB/s]
61%|██████ | 1.87G/3.08G [00:57<00:43, 28.2MiB/s]
61%|██████ | 1.88G/3.08G [00:57<00:36, 33.0MiB/s]
61%|██████ | 1.89G/3.08G [00:58<00:33, 36.3MiB/s]
61%|██████▏ | 1.89G/3.08G [00:58<00:37, 31.8MiB/s]
61%|██████▏ | 1.89G/3.08G [00:58<00:36, 32.5MiB/s]
62%|██████▏ | 1.90G/3.08G [00:58<00:38, 30.7MiB/s]
62%|██████▏ | 1.90G/3.08G [00:58<00:34, 33.9MiB/s]
62%|██████▏ | 1.91G/3.08G [00:58<00:38, 30.6MiB/s]
62%|██████▏ | 1.91G/3.08G [00:59<00:38, 30.2MiB/s]
62%|██████▏ | 1.92G/3.08G [00:59<00:30, 38.3MiB/s]
63%|██████▎ | 1.93G/3.08G [00:59<00:26, 44.2MiB/s]
63%|██████▎ | 1.93G/3.08G [00:59<00:34, 33.7MiB/s]
63%|██████▎ | 1.94G/3.08G [00:59<00:34, 32.8MiB/s]
63%|██████▎ | 1.94G/3.08G [00:59<00:33, 34.4MiB/s]
63%|██████▎ | 1.95G/3.08G [01:00<00:38, 29.6MiB/s]
63%|██████▎ | 1.95G/3.08G [01:00<00:35, 32.2MiB/s]
64%|██████▎ | 1.96G/3.08G [01:00<00:29, 37.8MiB/s]
64%|██████▎ | 1.97G/3.08G [01:00<00:33, 33.0MiB/s]
64%|██████▍ | 1.97G/3.08G [01:00<00:37, 29.9MiB/s]
64%|██████▍ | 1.97G/3.08G [01:00<00:41, 26.6MiB/s]
64%|██████▍ | 1.98G/3.08G [01:00<00:34, 31.6MiB/s]
64%|██████▍ | 1.98G/3.08G [01:01<00:37, 29.6MiB/s]
64%|██████▍ | 1.99G/3.08G [01:01<00:35, 31.2MiB/s]
65%|██████▍ | 2.00G/3.08G [01:01<00:26, 41.0MiB/s]
65%|██████▍ | 2.00G/3.08G [01:01<00:35, 30.8MiB/s]
65%|██████▍ | 2.00G/3.08G [01:01<00:43, 24.7MiB/s]
65%|██████▌ | 2.01G/3.08G [01:02<00:42, 25.4MiB/s]
65%|██████▌ | 2.01G/3.08G [01:02<00:41, 26.1MiB/s]
65%|██████▌ | 2.01G/3.08G [01:02<00:43, 24.5MiB/s]
65%|██████▌ | 2.02G/3.08G [01:02<00:35, 29.7MiB/s]
66%|██████▌ | 2.02G/3.08G [01:02<00:35, 29.8MiB/s]
66%|██████▌ | 2.03G/3.08G [01:02<00:37, 28.3MiB/s]
66%|██████▌ | 2.03G/3.08G [01:02<00:39, 26.9MiB/s]
66%|██████▌ | 2.03G/3.08G [01:02<00:42, 24.6MiB/s]
66%|██████▌ | 2.04G/3.08G [01:03<00:31, 32.7MiB/s]
66%|██████▋ | 2.05G/3.08G [01:03<00:23, 44.1MiB/s]
67%|██████▋ | 2.05G/3.08G [01:03<00:27, 37.0MiB/s]
67%|██████▋ | 2.06G/3.08G [01:03<00:29, 34.4MiB/s]
67%|██████▋ | 2.06G/3.08G [01:03<00:33, 31.0MiB/s]
67%|██████▋ | 2.06G/3.08G [01:03<00:32, 31.1MiB/s]
67%|██████▋ | 2.07G/3.08G [01:03<00:39, 26.0MiB/s]
67%|██████▋ | 2.07G/3.08G [01:04<00:35, 28.7MiB/s]
67%|██████▋ | 2.07G/3.08G [01:04<00:36, 27.3MiB/s]
67%|██████▋ | 2.08G/3.08G [01:04<00:27, 36.4MiB/s]
68%|██████▊ | 2.09G/3.08G [01:04<00:30, 32.9MiB/s]
68%|██████▊ | 2.09G/3.08G [01:04<00:30, 32.4MiB/s]
68%|██████▊ | 2.10G/3.08G [01:04<00:29, 33.1MiB/s]
68%|██████▊ | 2.10G/3.08G [01:04<00:31, 30.9MiB/s]
68%|██████▊ | 2.10G/3.08G [01:05<00:31, 31.3MiB/s]
68%|██████▊ | 2.11G/3.08G [01:05<00:33, 29.3MiB/s]
68%|██████▊ | 2.11G/3.08G [01:05<00:36, 26.3MiB/s]
69%|██████▊ | 2.11G/3.08G [01:05<00:38, 25.2MiB/s]
69%|██████▉ | 2.12G/3.08G [01:05<00:34, 28.1MiB/s]
69%|██████▉ | 2.13G/3.08G [01:05<00:25, 38.1MiB/s]
69%|██████▉ | 2.13G/3.08G [01:06<00:26, 35.4MiB/s]
69%|██████▉ | 2.14G/3.08G [01:06<00:32, 29.5MiB/s]
69%|██████▉ | 2.14G/3.08G [01:06<00:33, 28.3MiB/s]
70%|██████▉ | 2.15G/3.08G [01:06<00:30, 30.6MiB/s]
70%|██████▉ | 2.15G/3.08G [01:06<00:25, 36.2MiB/s]
70%|██████▉ | 2.16G/3.08G [01:06<00:30, 30.8MiB/s]
70%|███████ | 2.16G/3.08G [01:07<00:28, 32.1MiB/s]
70%|███████ | 2.17G/3.08G [01:07<00:27, 32.7MiB/s]
71%|███████ | 2.17G/3.08G [01:07<00:30, 29.4MiB/s]
71%|███████ | 2.18G/3.08G [01:07<00:29, 31.1MiB/s]
71%|███████ | 2.18G/3.08G [01:07<00:34, 25.9MiB/s]
71%|███████ | 2.19G/3.08G [01:07<00:30, 29.8MiB/s]
71%|███████ | 2.19G/3.08G [01:08<00:39, 22.7MiB/s]
71%|███████ | 2.20G/3.08G [01:08<00:31, 28.1MiB/s]
71%|███████▏ | 2.20G/3.08G [01:08<00:34, 25.5MiB/s]
72%|███████▏ | 2.21G/3.08G [01:08<00:28, 30.4MiB/s]
72%|███████▏ | 2.21G/3.08G [01:08<00:22, 37.9MiB/s]
72%|███████▏ | 2.22G/3.08G [01:08<00:20, 42.5MiB/s]
72%|███████▏ | 2.23G/3.08G [01:09<00:24, 35.3MiB/s]
72%|███████▏ | 2.23G/3.08G [01:09<00:23, 36.8MiB/s]
73%|███████▎ | 2.24G/3.08G [01:09<00:19, 43.1MiB/s]
73%|███████▎ | 2.24G/3.08G [01:09<00:20, 40.1MiB/s]
73%|███████▎ | 2.25G/3.08G [01:09<00:22, 37.3MiB/s]
73%|███████▎ | 2.25G/3.08G [01:09<00:22, 36.4MiB/s]
73%|███████▎ | 2.26G/3.08G [01:09<00:28, 29.3MiB/s]
73%|███████▎ | 2.26G/3.08G [01:10<00:27, 29.7MiB/s]
73%|███████▎ | 2.27G/3.08G [01:10<00:30, 26.5MiB/s]
74%|███████▎ | 2.27G/3.08G [01:10<00:27, 29.1MiB/s]
74%|███████▎ | 2.27G/3.08G [01:10<00:29, 27.3MiB/s]
74%|███████▍ | 2.28G/3.08G [01:10<00:25, 31.2MiB/s]
74%|███████▍ | 2.28G/3.08G [01:10<00:27, 29.1MiB/s]
74%|███████▍ | 2.29G/3.08G [01:10<00:25, 31.2MiB/s]
74%|███████▍ | 2.29G/3.08G [01:11<00:27, 29.0MiB/s]
75%|███████▍ | 2.30G/3.08G [01:11<00:20, 37.5MiB/s]
75%|███████▍ | 2.30G/3.08G [01:11<00:22, 35.0MiB/s]
75%|███████▍ | 2.31G/3.08G [01:11<00:25, 30.3MiB/s]
75%|███████▍ | 2.31G/3.08G [01:11<00:26, 29.1MiB/s]
75%|███████▌ | 2.31G/3.08G [01:11<00:21, 35.4MiB/s]
75%|███████▌ | 2.32G/3.08G [01:11<00:25, 29.6MiB/s]
75%|███████▌ | 2.32G/3.08G [01:12<00:25, 30.5MiB/s]
75%|███████▌ | 2.33G/3.08G [01:12<00:27, 27.2MiB/s]
76%|███████▌ | 2.33G/3.08G [01:12<00:25, 29.8MiB/s]
76%|███████▌ | 2.34G/3.08G [01:12<00:22, 33.5MiB/s]
76%|███████▌ | 2.35G/3.08G [01:12<00:18, 39.8MiB/s]
76%|███████▋ | 2.36G/3.08G [01:12<00:18, 38.6MiB/s]
77%|███████▋ | 2.36G/3.08G [01:13<00:19, 36.9MiB/s]
77%|███████▋ | 2.37G/3.08G [01:13<00:20, 35.7MiB/s]
77%|███████▋ | 2.37G/3.08G [01:13<00:17, 41.0MiB/s]
77%|███████▋ | 2.38G/3.08G [01:13<00:18, 39.1MiB/s]
77%|███████▋ | 2.38G/3.08G [01:13<00:21, 32.2MiB/s]
77%|███████▋ | 2.39G/3.08G [01:13<00:22, 31.5MiB/s]
78%|███████▊ | 2.39G/3.08G [01:14<00:23, 29.3MiB/s]
78%|███████▊ | 2.40G/3.08G [01:14<00:19, 35.9MiB/s]
78%|███████▊ | 2.41G/3.08G [01:14<00:16, 41.6MiB/s]
78%|███████▊ | 2.41G/3.08G [01:14<00:17, 38.3MiB/s]
78%|███████▊ | 2.42G/3.08G [01:14<00:24, 27.2MiB/s]
79%|███████▊ | 2.42G/3.08G [01:15<00:20, 31.5MiB/s]
79%|███████▉ | 2.43G/3.08G [01:15<00:16, 38.5MiB/s]
79%|███████▉ | 2.44G/3.08G [01:15<00:15, 41.2MiB/s]
79%|███████▉ | 2.45G/3.08G [01:15<00:15, 41.4MiB/s]
80%|███████▉ | 2.45G/3.08G [01:15<00:19, 33.2MiB/s]
80%|███████▉ | 2.46G/3.08G [01:15<00:19, 31.8MiB/s]
80%|███████▉ | 2.46G/3.08G [01:16<00:16, 38.3MiB/s]
80%|████████ | 2.47G/3.08G [01:16<00:17, 34.9MiB/s]
80%|████████ | 2.47G/3.08G [01:16<00:18, 33.8MiB/s]
81%|████████ | 2.48G/3.08G [01:16<00:14, 41.2MiB/s]
81%|████████ | 2.49G/3.08G [01:16<00:13, 45.1MiB/s]
81%|████████ | 2.49G/3.08G [01:16<00:14, 40.5MiB/s]
81%|████████ | 2.50G/3.08G [01:16<00:16, 35.6MiB/s]
81%|████████▏ | 2.51G/3.08G [01:17<00:14, 40.6MiB/s]
81%|████████▏ | 2.51G/3.08G [01:17<00:14, 38.3MiB/s]
82%|████████▏ | 2.52G/3.08G [01:17<00:15, 37.6MiB/s]
82%|████████▏ | 2.52G/3.08G [01:17<00:16, 34.6MiB/s]
82%|████████▏ | 2.52G/3.08G [01:17<00:14, 38.2MiB/s]
82%|████████▏ | 2.53G/3.08G [01:17<00:17, 31.9MiB/s]
82%|████████▏ | 2.53G/3.08G [01:17<00:16, 34.0MiB/s]
82%|████████▏ | 2.54G/3.08G [01:18<00:12, 42.0MiB/s]
83%|████████▎ | 2.54G/3.08G [01:18<00:17, 31.0MiB/s]
83%|████████▎ | 2.55G/3.08G [01:18<00:16, 31.5MiB/s]
83%|████████▎ | 2.56G/3.08G [01:18<00:14, 37.5MiB/s]
83%|████████▎ | 2.56G/3.08G [01:18<00:23, 22.6MiB/s]
83%|████████▎ | 2.57G/3.08G [01:19<00:18, 27.5MiB/s]
83%|████████▎ | 2.57G/3.08G [01:19<00:18, 27.2MiB/s]
84%|████████▎ | 2.58G/3.08G [01:19<00:19, 26.5MiB/s]
84%|████████▎ | 2.58G/3.08G [01:19<00:15, 32.1MiB/s]
84%|████████▍ | 2.59G/3.08G [01:19<00:19, 25.4MiB/s]
84%|████████▍ | 2.59G/3.08G [01:19<00:17, 27.9MiB/s]
84%|████████▍ | 2.59G/3.08G [01:20<00:19, 25.3MiB/s]
84%|████████▍ | 2.60G/3.08G [01:20<00:16, 29.3MiB/s]
84%|████████▍ | 2.60G/3.08G [01:20<00:23, 20.9MiB/s]
85%|████████▍ | 2.61G/3.08G [01:20<00:18, 26.4MiB/s]
85%|████████▍ | 2.61G/3.08G [01:20<00:18, 26.2MiB/s]
85%|████████▍ | 2.62G/3.08G [01:20<00:16, 29.2MiB/s]
85%|████████▍ | 2.62G/3.08G [01:21<00:17, 27.0MiB/s]
85%|████████▌ | 2.62G/3.08G [01:21<00:14, 30.7MiB/s]
85%|████████▌ | 2.63G/3.08G [01:21<00:15, 29.7MiB/s]
85%|████████▌ | 2.63G/3.08G [01:21<00:11, 38.7MiB/s]
86%|████████▌ | 2.64G/3.08G [01:21<00:13, 33.9MiB/s]
86%|████████▌ | 2.64G/3.08G [01:21<00:14, 29.5MiB/s]
86%|████████▌ | 2.65G/3.08G [01:21<00:12, 35.7MiB/s]
86%|████████▌ | 2.65G/3.08G [01:22<00:14, 30.6MiB/s]
86%|████████▌ | 2.66G/3.08G [01:22<00:14, 29.1MiB/s]
86%|████████▋ | 2.67G/3.08G [01:22<00:10, 39.2MiB/s]
87%|████████▋ | 2.67G/3.08G [01:22<00:09, 42.3MiB/s]
87%|████████▋ | 2.68G/3.08G [01:22<00:13, 29.3MiB/s]
87%|████████▋ | 2.68G/3.08G [01:22<00:12, 31.9MiB/s]
87%|████████▋ | 2.69G/3.08G [01:23<00:12, 31.7MiB/s]
87%|████████▋ | 2.69G/3.08G [01:23<00:12, 30.0MiB/s]
88%|████████▊ | 2.70G/3.08G [01:23<00:12, 30.9MiB/s]
88%|████████▊ | 2.70G/3.08G [01:23<00:12, 30.5MiB/s]
88%|████████▊ | 2.71G/3.08G [01:23<00:12, 31.3MiB/s]
88%|████████▊ | 2.71G/3.08G [01:23<00:15, 24.5MiB/s]
88%|████████▊ | 2.72G/3.08G [01:24<00:13, 27.5MiB/s]
88%|████████▊ | 2.72G/3.08G [01:24<00:15, 23.7MiB/s]
88%|████████▊ | 2.72G/3.08G [01:24<00:12, 28.2MiB/s]
88%|████████▊ | 2.73G/3.08G [01:24<00:11, 29.9MiB/s]
89%|████████▊ | 2.73G/3.08G [01:24<00:11, 29.7MiB/s]
89%|████████▉ | 2.74G/3.08G [01:24<00:08, 40.4MiB/s]
89%|████████▉ | 2.75G/3.08G [01:25<00:09, 36.3MiB/s]
89%|████████▉ | 2.75G/3.08G [01:25<00:10, 32.6MiB/s]
90%|████████▉ | 2.76G/3.08G [01:25<00:07, 41.5MiB/s]
90%|████████▉ | 2.77G/3.08G [01:25<00:08, 36.8MiB/s]
90%|████████▉ | 2.77G/3.08G [01:25<00:10, 30.1MiB/s]
90%|█████████ | 2.78G/3.08G [01:25<00:09, 31.3MiB/s]
90%|█████████ | 2.78G/3.08G [01:26<00:09, 33.3MiB/s]
90%|█████████ | 2.79G/3.08G [01:26<00:10, 29.0MiB/s]
91%|█████████ | 2.79G/3.08G [01:26<00:09, 30.7MiB/s]
91%|█████████ | 2.80G/3.08G [01:26<00:09, 29.7MiB/s]
91%|█████████ | 2.80G/3.08G [01:26<00:08, 33.9MiB/s]
91%|█████████ | 2.81G/3.08G [01:26<00:06, 39.9MiB/s]
91%|█████████▏| 2.82G/3.08G [01:27<00:07, 34.9MiB/s]
91%|█████████▏| 2.82G/3.08G [01:27<00:08, 30.5MiB/s]
92%|█████████▏| 2.83G/3.08G [01:27<00:08, 30.9MiB/s]
92%|█████████▏| 2.83G/3.08G [01:27<00:07, 32.0MiB/s]
92%|█████████▏| 2.84G/3.08G [01:27<00:08, 30.1MiB/s]
92%|█████████▏| 2.84G/3.08G [01:27<00:06, 34.9MiB/s]
92%|█████████▏| 2.85G/3.08G [01:28<00:07, 31.2MiB/s]
92%|█████████▏| 2.85G/3.08G [01:28<00:07, 30.8MiB/s]
93%|█████████▎| 2.86G/3.08G [01:28<00:06, 36.7MiB/s]
93%|█████████▎| 2.86G/3.08G [01:28<00:06, 34.6MiB/s]
93%|█████████▎| 2.87G/3.08G [01:28<00:05, 36.2MiB/s]
93%|█████████▎| 2.88G/3.08G [01:28<00:05, 36.1MiB/s]
93%|█████████▎| 2.88G/3.08G [01:28<00:05, 35.8MiB/s]
94%|█████████▎| 2.89G/3.08G [01:29<00:05, 33.4MiB/s]
94%|█████████▍| 2.89G/3.08G [01:29<00:05, 37.0MiB/s]
94%|█████████▍| 2.90G/3.08G [01:29<00:04, 38.1MiB/s]
94%|█████████▍| 2.90G/3.08G [01:29<00:04, 37.7MiB/s]
94%|█████████▍| 2.91G/3.08G [01:29<00:04, 43.4MiB/s]
94%|█████████▍| 2.91G/3.08G [01:29<00:04, 38.8MiB/s]
95%|█████████▍| 2.92G/3.08G [01:29<00:04, 37.2MiB/s]
95%|█████████▍| 2.92G/3.08G [01:30<00:04, 33.0MiB/s]
95%|█████████▍| 2.93G/3.08G [01:30<00:04, 38.8MiB/s]
95%|█████████▌| 2.94G/3.08G [01:30<00:03, 39.5MiB/s]
95%|█████████▌| 2.94G/3.08G [01:30<00:03, 46.3MiB/s]
96%|█████████▌| 2.95G/3.08G [01:30<00:02, 47.6MiB/s]
96%|█████████▌| 2.96G/3.08G [01:30<00:02, 50.6MiB/s]
96%|█████████▌| 2.96G/3.08G [01:31<00:02, 39.9MiB/s]
96%|█████████▋| 2.97G/3.08G [01:31<00:03, 37.2MiB/s]
96%|█████████▋| 2.97G/3.08G [01:31<00:03, 35.7MiB/s]
97%|█████████▋| 2.98G/3.08G [01:31<00:03, 33.7MiB/s]
97%|█████████▋| 2.98G/3.08G [01:31<00:03, 30.1MiB/s]
97%|█████████▋| 2.99G/3.08G [01:31<00:03, 27.3MiB/s]
97%|█████████▋| 2.99G/3.08G [01:32<00:03, 28.8MiB/s]
97%|█████████▋| 3.00G/3.08G [01:32<00:03, 24.8MiB/s]
97%|█████████▋| 3.00G/3.08G [01:32<00:02, 27.7MiB/s]
98%|█████████▊| 3.01G/3.08G [01:32<00:02, 34.2MiB/s]
98%|█████████▊| 3.02G/3.08G [01:32<00:01, 33.2MiB/s]
98%|█████████▊| 3.02G/3.08G [01:33<00:02, 29.5MiB/s]
98%|█████████▊| 3.03G/3.08G [01:33<00:01, 31.2MiB/s]
98%|█████████▊| 3.04G/3.08G [01:33<00:01, 32.5MiB/s]
99%|█████████▊| 3.05G/3.08G [01:33<00:00, 40.4MiB/s]
99%|█████████▉| 3.05G/3.08G [01:33<00:00, 46.9MiB/s]
99%|█████████▉| 3.06G/3.08G [01:33<00:00, 42.0MiB/s]
99%|█████████▉| 3.06G/3.08G [01:34<00:00, 35.2MiB/s]
100%|█████████▉| 3.07G/3.08G [01:34<00:00, 40.7MiB/s]
100%|█████████▉| 3.07G/3.08G [01:34<00:00, 40.1MiB/s]
100%|█████████▉| 3.08G/3.08G [01:34<00:00, 37.3MiB/s]
100%|██████████| 3.08G/3.08G [01:34<00:00, 32.6MiB/s]
然后我们可以将标注加载到数据框中。
dataset_path = os.path.join(download_dir, 'Stanford_Online_Products')
train_data = pd.read_csv(f'{dataset_path}/train.csv', index_col=0)
test_data = pd.read_csv(f'{dataset_path}/test.csv', index_col=0)
image_col_1 = "Image1"
image_col_2 = "Image2"
label_col = "Label"
match_label = 1
在这里，您需要指定 match_label，即代表图像对语义匹配的标签类别。在此演示数据集中，我们使用 1，因为我们将同一产品的图像对标记为 1。您可以根据您的任务上下文来指定 match_label。
接下来，我们扩展图像路径，因为原始路径是相对路径。
def path_expander(path, base_folder):
path_l = path.split(';')
return ';'.join([os.path.abspath(os.path.join(base_folder, path)) for path in path_l])
for image_col in [image_col_1, image_col_2]:
train_data[image_col] = train_data[image_col].apply(lambda ele: path_expander(ele, base_folder=dataset_path))
test_data[image_col] = test_data[image_col].apply(lambda ele: path_expander(ele, base_folder=dataset_path))
标注只包含图像路径对及其二分类标签（1 和 0 分别表示图像对匹配或不匹配）。
train_data.head()
| 图像 1 | 图像 2 | 标签 | |
|---|---|---|---|
| 0 | /home/ci/autogluon/docs/tutorials/multimodal/s... | /home/ci/autogluon/docs/tutorials/multimodal/s... | 0 |
| 1 | /home/ci/autogluon/docs/tutorials/multimodal/s... | /home/ci/autogluon/docs/tutorials/multimodal/s... | 1 |
| 2 | /home/ci/autogluon/docs/tutorials/multimodal/s... | /home/ci/autogluon/docs/tutorials/multimodal/s... | 0 |
| 3 | /home/ci/autogluon/docs/tutorials/multimodal/s... | /home/ci/autogluon/docs/tutorials/multimodal/s... | 1 |
| 4 | /home/ci/autogluon/docs/tutorials/multimodal/s... | /home/ci/autogluon/docs/tutorials/multimodal/s... | 1 |
让我们可视化一个匹配的图像对。
pil_img = Image(filename=train_data[image_col_1][5])
display(pil_img)
pil_img = Image(filename=train_data[image_col_2][5])
display(pil_img)
这是两个不匹配的图像。
pil_img = Image(filename=train_data[image_col_1][0])
display(pil_img)
pil_img = Image(filename=train_data[image_col_2][0])
display(pil_img)
训练模型¶
理想情况下，我们希望获得一个能够为正样本对/负样本对返回高分/低分的模型。使用 AutoMM，我们可以轻松训练一个捕获图像之间语义关系的模型。基本上，它使用 Swin Transformer 将每张图像投影到高维向量，并计算特征向量的余弦相似度。
使用 AutoMM，您只需指定 query、response 和 label 列名，然后在训练数据集上拟合模型，无需担心实现细节。
from autogluon.multimodal import MultiModalPredictor
predictor = MultiModalPredictor(
problem_type="image_similarity",
query=image_col_1, # the column name of the first image
response=image_col_2, # the column name of the second image
label=label_col, # the label column name
match_label=match_label, # the label indicating that query and response have the same semantic meanings.
eval_metric='auc', # the evaluation metric
)
# Fit the model
predictor.fit(
train_data=train_data,
time_limit=180,
)
No path specified. Models will be saved in: "AutogluonModels/ag-20250508_205823"
=================== System Info ===================
AutoGluon Version: 1.3.1b20250508
Python Version: 3.11.9
Operating System: Linux
Platform Machine: x86_64
Platform Version: #1 SMP Wed Mar 12 14:53:59 UTC 2025
CPU Count: 8
Pytorch Version: 2.6.0+cu124
CUDA Version: 12.4
Memory Avail: 28.20 GB / 30.95 GB (91.1%)
Disk Space Avail: 178.56 GB / 255.99 GB (69.8%)
===================================================
AutoGluon infers your prediction problem is: 'binary' (because only two unique label-values observed).
2 unique label values: [np.int64(0), np.int64(1)]
If 'binary' is not the correct problem_type, please manually specify the problem_type parameter during Predictor init (You may specify problem_type as one of: ['binary', 'multiclass', 'regression', 'quantile'])
AutoMM starts to create your model. ✨✨✨
To track the learning progress, you can open a terminal and launch Tensorboard:
```shell
# Assume you have installed tensorboard
tensorboard --logdir /home/ci/autogluon/docs/tutorials/multimodal/semantic_matching/AutogluonModels/ag-20250508_205823
```
INFO: Seed set to 0
WARNING:timm.models._builder:Unexpected keys (head.fc.fc1.bias, head.fc.fc1.weight, head.fc.norm.bias, head.fc.norm.weight) found while loading pretrained weights. This may be expected if model is being adapted.
GPU Count: 1
GPU Count to be Used: 1
INFO: Using 16bit Automatic Mixed Precision (AMP)
INFO: GPU available: True (cuda), used: True
INFO: TPU available: False, using: 0 TPU cores
INFO: HPU available: False, using: 0 HPUs
INFO: LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
INFO:
| Name | Type | Params | Mode
------------------------------------------------------------------------------
0 | query_model | TimmAutoModelForImagePrediction | 93.3 M | train
1 | response_model | TimmAutoModelForImagePrediction | 93.3 M | train
2 | validation_metric | BinaryAUROC | 0 | train
3 | loss_func | ContrastiveLoss | 0 | train
4 | miner_func | PairMarginMiner | 0 | train
------------------------------------------------------------------------------
93.3 M Trainable params
0 Non-trainable params
93.3 M Total params
373.248 Total estimated model params size (MB)
866 Modules in train mode
0 Modules in eval mode
INFO: Epoch 0, global step 15: 'val_roc_auc' reached 0.82784 (best 0.82784), saving model to '/home/ci/autogluon/docs/tutorials/multimodal/semantic_matching/AutogluonModels/ag-20250508_205823/epoch=0-step=15.ckpt' as top 3
INFO: Time limit reached. Elapsed time is 0:03:00. Signaling Trainer to stop.
INFO: Epoch 0, global step 23: 'val_roc_auc' reached 0.89310 (best 0.89310), saving model to '/home/ci/autogluon/docs/tutorials/multimodal/semantic_matching/AutogluonModels/ag-20250508_205823/epoch=0-step=23.ckpt' as top 3
Start to fuse 2 checkpoints via the greedy soup algorithm.
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
AutoMM has created your model. 🎉🎉🎉
To load the model, use the code below:
```python
from autogluon.multimodal import MultiModalPredictor
predictor = MultiModalPredictor.load("/home/ci/autogluon/docs/tutorials/multimodal/semantic_matching/AutogluonModels/ag-20250508_205823")
```
If you are not satisfied with the model, try to increase the training time,
adjust the hyperparameters (https://autogluon.cn/stable/tutorials/multimodal/advanced_topics/customization.html),
or post issues on GitHub (https://github.com/autogluon/autogluon/issues).
<autogluon.multimodal.predictor.MultiModalPredictor at 0x7f3b774fb010>
在测试数据集上评估¶
您可以在测试数据集上评估预测器，查看其在 roc_auc 分数上的表现。
score = predictor.evaluate(test_data)
print("evaluation score: ", score)
evaluation score: {'roc_auc': np.float64(0.8946748085592974)}
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
对图像对进行预测¶
给出新的图像对，我们可以预测它们是否匹配。
pred = predictor.predict(test_data.head(3))
print(pred)
0 1
1 1
2 1
Name: Label, dtype: int64
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
预测使用朴素的概率阈值 0.5。也就是说，我们选择概率大于 0.5 的标签。
预测匹配概率¶
然而，您可以通过获取概率来进行更自定义的阈值处理。
proba = predictor.predict_proba(test_data.head(3))
print(proba)
0 1
0 0.322680 0.677320
1 0.051566 0.948434
2 0.073345 0.926655
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
提取嵌入¶
您还可以提取图像对中每张图像的嵌入。
embeddings_1 = predictor.extract_embedding({image_col_1: test_data[image_col_1][:5].tolist()})
print(embeddings_1.shape)
embeddings_2 = predictor.extract_embedding({image_col_2: test_data[image_col_2][:5].tolist()})
print(embeddings_2.shape)
(5, 768)
(5, 768)
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
INFO: Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.
其他示例¶
您可以前往 AutoMM Examples 探索关于 AutoMM 的其他示例。
自定义¶
要了解如何自定义 AutoMM，请参考 自定义 AutoMM。