Image-Text Semantic Matching with AutoMM¶
Vision and language are two important aspects of human intelligence to understand the real world. Image-text semantic matching, measuring the visual-semantic similarity between image and text, plays a critical role in bridging the vision and language. Learning a joint space where text and image feature vectors are aligned is a typical solution for image-text matching. It is becoming increasingly significant for various vision-and-language tasks, such as cross-modal retrieval, image captioning, text-to-image synthesis, and multimodal neural machine translation. This tutorial will introduce how to apply AutoMM to the image-text matching task.
import os
import warnings
from IPython.display import Image, display
import numpy as np
warnings.filterwarnings('ignore')
np.random.seed(123)
Dataset¶
In this tutorial, we will use the Flickr30K dataset to demonstrate the image-text matching. The Flickr30k dataset is a popular benchmark for sentence-based picture portrayal. The dataset is comprised of 31,783 images that capture people engaged in everyday activities and events. Each image has a descriptive caption. We organized the dataset using pandas dataframe. To get started, Let’s download the dataset.
from autogluon.core.utils.loaders import load_pd
import pandas as pd
download_dir = './ag_automm_tutorial_imgtxt'
zip_file = 'https://automl-mm-bench.s3.amazonaws.com/flickr30k.zip'
from autogluon.core.utils.loaders import load_zip
load_zip.unzip(zip_file, unzip_dir=download_dir)
Downloading ./ag_automm_tutorial_imgtxt/file.zip from https://automl-mm-bench.s3.amazonaws.com/flickr30k.zip...
0%| | 0.00/4.38G [00:00<?, ?iB/s]
0%| | 4.68M/4.38G [00:00<01:33, 46.8MiB/s]
0%| | 9.35M/4.38G [00:00<04:32, 16.1MiB/s]
0%| | 15.4M/4.38G [00:00<02:58, 24.5MiB/s]
0%| | 19.0M/4.38G [00:00<02:55, 24.8MiB/s]
1%| | 23.4M/4.38G [00:00<02:42, 26.9MiB/s]
1%| | 26.6M/4.38G [00:01<03:12, 22.7MiB/s]
1%| | 31.8M/4.38G [00:01<02:47, 25.9MiB/s]
1%| | 34.7M/4.38G [00:01<03:07, 23.1MiB/s]
1%| | 40.2M/4.38G [00:01<02:37, 27.6MiB/s]
1%| | 43.1M/4.38G [00:01<03:17, 22.0MiB/s]
1%| | 47.4M/4.38G [00:02<03:21, 21.5MiB/s]
1%| | 50.3M/4.38G [00:02<03:42, 19.4MiB/s]
1%|▏ | 57.0M/4.38G [00:02<02:35, 27.8MiB/s]
1%|▏ | 60.3M/4.38G [00:02<03:10, 22.7MiB/s]
2%|▏ | 67.1M/4.38G [00:02<02:27, 29.3MiB/s]
2%|▏ | 75.0M/4.38G [00:02<01:49, 39.2MiB/s]
2%|▏ | 79.7M/4.38G [00:02<01:49, 39.3MiB/s]
2%|▏ | 84.3M/4.38G [00:03<02:40, 26.8MiB/s]
2%|▏ | 90.5M/4.38G [00:03<02:19, 30.9MiB/s]
2%|▏ | 94.3M/4.38G [00:03<02:32, 28.2MiB/s]
2%|▏ | 98.9M/4.38G [00:03<02:26, 29.3MiB/s]
2%|▏ | 102M/4.38G [00:03<03:08, 22.7MiB/s]
2%|▏ | 107M/4.38G [00:04<02:34, 27.7MiB/s]
3%|▎ | 111M/4.38G [00:04<02:37, 27.1MiB/s]
3%|▎ | 116M/4.38G [00:04<02:36, 27.2MiB/s]
3%|▎ | 119M/4.38G [00:04<02:55, 24.3MiB/s]
3%|▎ | 124M/4.38G [00:04<02:39, 26.7MiB/s]
3%|▎ | 127M/4.38G [00:04<03:06, 22.8MiB/s]
3%|▎ | 132M/4.38G [00:05<02:35, 27.3MiB/s]
3%|▎ | 135M/4.38G [00:05<02:59, 23.7MiB/s]
3%|▎ | 143M/4.38G [00:05<02:38, 26.8MiB/s]
3%|▎ | 149M/4.38G [00:05<03:00, 23.4MiB/s]
3%|▎ | 152M/4.38G [00:06<03:58, 17.7MiB/s]
4%|▎ | 158M/4.38G [00:06<03:38, 19.3MiB/s]
4%|▎ | 160M/4.38G [00:06<03:53, 18.1MiB/s]
4%|▍ | 166M/4.38G [00:06<03:31, 19.9MiB/s]
4%|▍ | 168M/4.38G [00:06<04:03, 17.3MiB/s]
4%|▍ | 174M/4.38G [00:07<03:19, 21.1MiB/s]
4%|▍ | 177M/4.38G [00:07<03:37, 19.3MiB/s]
4%|▍ | 185M/4.38G [00:07<02:47, 25.0MiB/s]
4%|▍ | 191M/4.38G [00:07<02:19, 30.1MiB/s]
4%|▍ | 194M/4.38G [00:07<02:58, 23.5MiB/s]
5%|▍ | 200M/4.38G [00:08<03:23, 20.6MiB/s]
5%|▍ | 202M/4.38G [00:08<03:34, 19.5MiB/s]
5%|▍ | 208M/4.38G [00:08<02:40, 26.0MiB/s]
5%|▍ | 211M/4.38G [00:08<02:53, 24.0MiB/s]
5%|▍ | 216M/4.38G [00:08<02:27, 28.3MiB/s]
5%|▌ | 220M/4.38G [00:08<02:30, 27.7MiB/s]
5%|▌ | 224M/4.38G [00:09<02:11, 31.7MiB/s]
5%|▌ | 228M/4.38G [00:09<02:51, 24.3MiB/s]
5%|▌ | 235M/4.38G [00:09<02:17, 30.0MiB/s]
5%|▌ | 238M/4.38G [00:09<02:18, 29.8MiB/s]
6%|▌ | 242M/4.38G [00:09<02:38, 26.2MiB/s]
6%|▌ | 245M/4.38G [00:10<03:48, 18.1MiB/s]
6%|▌ | 252M/4.38G [00:10<02:34, 26.7MiB/s]
6%|▌ | 255M/4.38G [00:10<02:52, 23.9MiB/s]
6%|▌ | 258M/4.38G [00:10<02:51, 24.0MiB/s]
6%|▌ | 261M/4.38G [00:10<03:02, 22.6MiB/s]
6%|▌ | 264M/4.38G [00:10<03:24, 20.1MiB/s]
6%|▌ | 268M/4.38G [00:11<03:05, 22.2MiB/s]
6%|▋ | 275M/4.38G [00:11<03:45, 18.2MiB/s]
6%|▋ | 278M/4.38G [00:11<03:44, 18.3MiB/s]
7%|▋ | 285M/4.38G [00:11<02:48, 24.3MiB/s]
7%|▋ | 293M/4.38G [00:11<01:59, 34.2MiB/s]
7%|▋ | 298M/4.38G [00:12<02:19, 29.3MiB/s]
7%|▋ | 301M/4.38G [00:12<02:17, 29.7MiB/s]
7%|▋ | 305M/4.38G [00:12<03:04, 22.1MiB/s]
7%|▋ | 309M/4.38G [00:12<02:59, 22.7MiB/s]
7%|▋ | 311M/4.38G [00:12<03:13, 21.0MiB/s]
7%|▋ | 317M/4.38G [00:12<02:36, 25.9MiB/s]
7%|▋ | 320M/4.38G [00:13<02:49, 23.9MiB/s]
7%|▋ | 325M/4.38G [00:13<02:52, 23.5MiB/s]
7%|▋ | 328M/4.38G [00:13<03:18, 20.4MiB/s]
8%|▊ | 334M/4.38G [00:13<03:21, 20.0MiB/s]
8%|▊ | 336M/4.38G [00:14<04:06, 16.4MiB/s]
8%|▊ | 342M/4.38G [00:14<03:31, 19.1MiB/s]
8%|▊ | 344M/4.38G [00:14<03:34, 18.8MiB/s]
8%|▊ | 351M/4.38G [00:14<02:43, 24.7MiB/s]
8%|▊ | 353M/4.38G [00:14<03:01, 22.2MiB/s]
8%|▊ | 361M/4.38G [00:14<02:27, 27.2MiB/s]
8%|▊ | 367M/4.38G [00:15<02:01, 33.2MiB/s]
8%|▊ | 371M/4.38G [00:15<02:16, 29.4MiB/s]
9%|▊ | 376M/4.38G [00:15<02:47, 23.9MiB/s]
9%|▊ | 378M/4.38G [00:15<03:15, 20.4MiB/s]
9%|▉ | 384M/4.38G [00:16<03:01, 22.0MiB/s]
9%|▉ | 386M/4.38G [00:16<03:15, 20.4MiB/s]
9%|▉ | 389M/4.38G [00:16<03:48, 17.5MiB/s]
9%|▉ | 394M/4.38G [00:16<03:00, 22.0MiB/s]
9%|▉ | 401M/4.38G [00:16<02:25, 27.4MiB/s]
9%|▉ | 404M/4.38G [00:16<02:48, 23.5MiB/s]
9%|▉ | 409M/4.38G [00:17<02:44, 24.1MiB/s]
9%|▉ | 412M/4.38G [00:17<02:57, 22.4MiB/s]
10%|▉ | 418M/4.38G [00:17<02:42, 24.4MiB/s]
10%|▉ | 420M/4.38G [00:17<02:50, 23.3MiB/s]
10%|▉ | 428M/4.38G [00:17<02:41, 24.6MiB/s]
10%|▉ | 430M/4.38G [00:17<02:53, 22.8MiB/s]
10%|▉ | 436M/4.38G [00:18<02:40, 24.6MiB/s]
10%|█ | 443M/4.38G [00:18<02:29, 26.3MiB/s]
10%|█ | 445M/4.38G [00:18<02:40, 24.5MiB/s]
10%|█ | 451M/4.38G [00:18<02:24, 27.3MiB/s]
10%|█ | 454M/4.38G [00:18<03:04, 21.3MiB/s]
11%|█ | 461M/4.38G [00:19<02:29, 26.2MiB/s]
11%|█ | 470M/4.38G [00:19<01:48, 36.2MiB/s]
11%|█ | 474M/4.38G [00:19<01:46, 36.8MiB/s]
11%|█ | 478M/4.38G [00:19<02:35, 25.1MiB/s]
11%|█ | 481M/4.38G [00:19<02:38, 24.6MiB/s]
11%|█ | 485M/4.38G [00:19<02:34, 25.2MiB/s]
11%|█ | 488M/4.38G [00:20<02:45, 23.5MiB/s]
11%|█ | 490M/4.38G [00:20<02:55, 22.2MiB/s]
11%|█▏ | 494M/4.38G [00:20<02:30, 25.8MiB/s]
11%|█▏ | 497M/4.38G [00:20<03:05, 20.9MiB/s]
11%|█▏ | 503M/4.38G [00:20<02:24, 26.9MiB/s]
12%|█▏ | 510M/4.38G [00:20<02:08, 30.2MiB/s]
12%|█▏ | 513M/4.38G [00:21<02:34, 25.0MiB/s]
12%|█▏ | 520M/4.38G [00:21<02:14, 28.8MiB/s]
12%|█▏ | 527M/4.38G [00:21<01:56, 33.1MiB/s]
12%|█▏ | 530M/4.38G [00:21<02:31, 25.3MiB/s]
12%|█▏ | 533M/4.38G [00:22<03:38, 17.6MiB/s]
12%|█▏ | 535M/4.38G [00:22<03:44, 17.1MiB/s]
12%|█▏ | 537M/4.38G [00:22<03:53, 16.5MiB/s]
12%|█▏ | 545M/4.38G [00:22<02:30, 25.5MiB/s]
13%|█▎ | 552M/4.38G [00:22<01:59, 32.0MiB/s]
13%|█▎ | 556M/4.38G [00:22<02:35, 24.6MiB/s]
13%|█▎ | 560M/4.38G [00:23<02:24, 26.5MiB/s]
13%|█▎ | 563M/4.38G [00:23<02:40, 23.7MiB/s]
13%|█▎ | 570M/4.38G [00:23<02:27, 25.8MiB/s]
13%|█▎ | 579M/4.38G [00:23<01:46, 35.8MiB/s]
13%|█▎ | 583M/4.38G [00:23<02:21, 26.8MiB/s]
13%|█▎ | 586M/4.38G [00:23<02:19, 27.3MiB/s]
13%|█▎ | 590M/4.38G [00:24<02:19, 27.2MiB/s]
14%|█▎ | 594M/4.38G [00:24<02:21, 26.9MiB/s]
14%|█▎ | 597M/4.38G [00:24<02:32, 24.8MiB/s]
14%|█▍ | 603M/4.38G [00:24<02:18, 27.3MiB/s]
14%|█▍ | 605M/4.38G [00:24<02:31, 25.0MiB/s]
14%|█▍ | 611M/4.38G [00:25<03:13, 19.4MiB/s]
14%|█▍ | 613M/4.38G [00:25<03:31, 17.8MiB/s]
14%|█▍ | 619M/4.38G [00:25<02:51, 22.0MiB/s]
14%|█▍ | 621M/4.38G [00:25<03:15, 19.3MiB/s]
14%|█▍ | 627M/4.38G [00:25<02:49, 22.1MiB/s]
14%|█▍ | 630M/4.38G [00:26<03:14, 19.3MiB/s]
15%|█▍ | 638M/4.38G [00:26<02:34, 24.2MiB/s]
15%|█▍ | 644M/4.38G [00:26<02:22, 26.1MiB/s]
15%|█▍ | 647M/4.38G [00:26<02:34, 24.2MiB/s]
15%|█▍ | 654M/4.38G [00:26<02:02, 30.4MiB/s]
15%|█▌ | 661M/4.38G [00:26<01:50, 33.8MiB/s]
15%|█▌ | 664M/4.38G [00:27<02:00, 30.8MiB/s]
15%|█▌ | 669M/4.38G [00:27<01:49, 34.0MiB/s]
15%|█▌ | 673M/4.38G [00:27<02:07, 29.1MiB/s]
15%|█▌ | 676M/4.38G [00:27<02:15, 27.3MiB/s]
15%|█▌ | 679M/4.38G [00:27<02:26, 25.3MiB/s]
16%|█▌ | 681M/4.38G [00:27<02:56, 21.0MiB/s]
16%|█▌ | 686M/4.38G [00:28<02:39, 23.2MiB/s]
16%|█▌ | 690M/4.38G [00:28<02:17, 26.8MiB/s]
16%|█▌ | 696M/4.38G [00:28<01:54, 32.2MiB/s]
16%|█▌ | 704M/4.38G [00:28<01:24, 43.7MiB/s]
16%|█▌ | 709M/4.38G [00:28<01:43, 35.3MiB/s]
16%|█▋ | 713M/4.38G [00:28<02:25, 25.3MiB/s]
16%|█▋ | 720M/4.38G [00:29<01:52, 32.6MiB/s]
17%|█▋ | 724M/4.38G [00:29<01:58, 31.0MiB/s]
17%|█▋ | 730M/4.38G [00:29<01:45, 34.7MiB/s]
17%|█▋ | 736M/4.38G [00:29<01:30, 40.2MiB/s]
17%|█▋ | 740M/4.38G [00:29<02:00, 30.3MiB/s]
17%|█▋ | 747M/4.38G [00:29<02:04, 29.1MiB/s]
17%|█▋ | 753M/4.38G [00:30<01:51, 32.4MiB/s]
17%|█▋ | 757M/4.38G [00:30<01:53, 32.0MiB/s]
17%|█▋ | 762M/4.38G [00:30<02:25, 24.9MiB/s]
17%|█▋ | 765M/4.38G [00:30<02:39, 22.6MiB/s]
18%|█▊ | 770M/4.38G [00:30<02:25, 24.9MiB/s]
18%|█▊ | 773M/4.38G [00:31<02:50, 21.2MiB/s]
18%|█▊ | 778M/4.38G [00:31<03:17, 18.2MiB/s]
18%|█▊ | 780M/4.38G [00:31<04:33, 13.2MiB/s]
18%|█▊ | 787M/4.38G [00:31<03:28, 17.2MiB/s]
18%|█▊ | 789M/4.38G [00:32<03:51, 15.5MiB/s]
18%|█▊ | 795M/4.38G [00:32<04:37, 12.9MiB/s]
18%|█▊ | 797M/4.38G [00:32<04:37, 12.9MiB/s]
18%|█▊ | 804M/4.38G [00:33<03:12, 18.6MiB/s]
18%|█▊ | 806M/4.38G [00:33<03:27, 17.3MiB/s]
19%|█▊ | 812M/4.38G [00:33<02:38, 22.5MiB/s]
19%|█▊ | 815M/4.38G [00:33<03:10, 18.7MiB/s]
19%|█▊ | 821M/4.38G [00:33<02:32, 23.3MiB/s]
19%|█▉ | 823M/4.38G [00:33<02:31, 23.4MiB/s]
19%|█▉ | 830M/4.38G [00:34<01:59, 29.7MiB/s]
19%|█▉ | 837M/4.38G [00:34<01:50, 32.0MiB/s]
19%|█▉ | 840M/4.38G [00:34<02:29, 23.7MiB/s]
19%|█▉ | 847M/4.38G [00:34<02:17, 25.7MiB/s]
19%|█▉ | 854M/4.38G [00:34<02:01, 29.1MiB/s]
20%|█▉ | 857M/4.38G [00:35<02:02, 28.7MiB/s]
20%|█▉ | 863M/4.38G [00:35<01:45, 33.3MiB/s]
20%|█▉ | 866M/4.38G [00:35<02:06, 27.7MiB/s]
20%|█▉ | 871M/4.38G [00:35<02:08, 27.3MiB/s]
20%|█▉ | 874M/4.38G [00:35<02:17, 25.5MiB/s]
20%|██ | 877M/4.38G [00:35<02:31, 23.1MiB/s]
20%|██ | 881M/4.38G [00:36<02:43, 21.4MiB/s]
20%|██ | 887M/4.38G [00:36<02:50, 20.5MiB/s]
20%|██ | 890M/4.38G [00:36<03:18, 17.6MiB/s]
20%|██ | 898M/4.38G [00:36<02:34, 22.5MiB/s]
21%|██ | 904M/4.38G [00:37<02:21, 24.7MiB/s]
21%|██ | 907M/4.38G [00:37<02:38, 21.9MiB/s]
21%|██ | 913M/4.38G [00:37<02:02, 28.4MiB/s]
21%|██ | 916M/4.38G [00:37<02:43, 21.2MiB/s]
21%|██ | 922M/4.38G [00:37<02:08, 26.9MiB/s]
21%|██ | 925M/4.38G [00:37<02:12, 26.0MiB/s]
21%|██ | 929M/4.38G [00:38<02:34, 22.3MiB/s]
21%|██▏ | 932M/4.38G [00:38<03:00, 19.1MiB/s]
21%|██▏ | 938M/4.38G [00:38<02:43, 21.1MiB/s]
21%|██▏ | 940M/4.38G [00:38<02:47, 20.6MiB/s]
22%|██▏ | 945M/4.38G [00:38<02:20, 24.4MiB/s]
22%|██▏ | 948M/4.38G [00:39<02:27, 23.2MiB/s]
22%|██▏ | 955M/4.38G [00:39<01:49, 31.3MiB/s]
22%|██▏ | 958M/4.38G [00:39<01:54, 29.8MiB/s]
22%|██▏ | 965M/4.38G [00:39<01:42, 33.5MiB/s]
22%|██▏ | 968M/4.38G [00:39<01:44, 32.8MiB/s]
22%|██▏ | 972M/4.38G [00:39<02:00, 28.3MiB/s]
22%|██▏ | 975M/4.38G [00:39<02:22, 23.8MiB/s]
22%|██▏ | 980M/4.38G [00:40<02:25, 23.4MiB/s]
22%|██▏ | 982M/4.38G [00:40<02:45, 20.5MiB/s]
23%|██▎ | 989M/4.38G [00:40<01:57, 28.9MiB/s]
23%|██▎ | 992M/4.38G [00:40<02:32, 22.2MiB/s]
23%|██▎ | 997M/4.38G [00:40<02:06, 26.7MiB/s]
23%|██▎ | 1.00G/4.38G [00:40<02:32, 22.2MiB/s]
23%|██▎ | 1.01G/4.38G [00:41<02:22, 23.7MiB/s]
23%|██▎ | 1.01G/4.38G [00:41<02:52, 19.5MiB/s]
23%|██▎ | 1.01G/4.38G [00:41<03:17, 17.0MiB/s]
23%|██▎ | 1.02G/4.38G [00:41<02:51, 19.6MiB/s]
23%|██▎ | 1.02G/4.38G [00:41<02:18, 24.3MiB/s]
23%|██▎ | 1.02G/4.38G [00:42<02:44, 20.4MiB/s]
24%|██▎ | 1.03G/4.38G [00:42<02:12, 25.3MiB/s]
24%|██▎ | 1.04G/4.38G [00:42<01:59, 28.1MiB/s]
24%|██▎ | 1.04G/4.38G [00:42<02:19, 23.9MiB/s]
24%|██▍ | 1.05G/4.38G [00:42<02:01, 27.4MiB/s]
24%|██▍ | 1.05G/4.38G [00:43<02:32, 21.9MiB/s]
24%|██▍ | 1.06G/4.38G [00:43<02:25, 22.9MiB/s]
24%|██▍ | 1.06G/4.38G [00:43<02:41, 20.6MiB/s]
24%|██▍ | 1.06G/4.38G [00:43<02:07, 26.1MiB/s]
24%|██▍ | 1.07G/4.38G [00:43<02:23, 23.1MiB/s]
25%|██▍ | 1.07G/4.38G [00:44<01:58, 27.8MiB/s]
25%|██▍ | 1.08G/4.38G [00:44<01:56, 28.2MiB/s]
25%|██▍ | 1.08G/4.38G [00:44<02:02, 26.8MiB/s]
25%|██▍ | 1.09G/4.38G [00:44<01:48, 30.4MiB/s]
25%|██▍ | 1.09G/4.38G [00:44<01:55, 28.4MiB/s]
25%|██▌ | 1.10G/4.38G [00:45<02:27, 22.2MiB/s]
25%|██▌ | 1.10G/4.38G [00:45<02:51, 19.1MiB/s]
25%|██▌ | 1.11G/4.38G [00:45<01:53, 28.7MiB/s]
25%|██▌ | 1.12G/4.38G [00:45<01:24, 38.8MiB/s]
26%|██▌ | 1.12G/4.38G [00:45<01:31, 35.7MiB/s]
26%|██▌ | 1.12G/4.38G [00:45<01:57, 27.7MiB/s]
26%|██▌ | 1.13G/4.38G [00:45<01:44, 31.2MiB/s]
26%|██▌ | 1.13G/4.38G [00:46<01:59, 27.1MiB/s]
26%|██▌ | 1.14G/4.38G [00:46<01:47, 30.2MiB/s]
26%|██▌ | 1.14G/4.38G [00:46<02:06, 25.7MiB/s]
26%|██▌ | 1.15G/4.38G [00:46<02:06, 25.5MiB/s]
26%|██▋ | 1.15G/4.38G [00:46<02:03, 26.1MiB/s]
26%|██▋ | 1.16G/4.38G [00:46<01:41, 31.8MiB/s]
26%|██▋ | 1.16G/4.38G [00:47<01:45, 30.6MiB/s]
27%|██▋ | 1.17G/4.38G [00:47<01:36, 33.3MiB/s]
27%|██▋ | 1.17G/4.38G [00:47<01:46, 30.1MiB/s]
27%|██▋ | 1.18G/4.38G [00:47<01:57, 27.2MiB/s]
27%|██▋ | 1.18G/4.38G [00:47<01:51, 28.7MiB/s]
27%|██▋ | 1.19G/4.38G [00:48<01:48, 29.4MiB/s]
27%|██▋ | 1.19G/4.38G [00:48<02:02, 26.1MiB/s]
27%|██▋ | 1.20G/4.38G [00:48<01:55, 27.7MiB/s]
27%|██▋ | 1.20G/4.38G [00:48<01:59, 26.6MiB/s]
28%|██▊ | 1.21G/4.38G [00:48<01:35, 33.1MiB/s]
28%|██▊ | 1.21G/4.38G [00:48<01:25, 36.9MiB/s]
28%|██▊ | 1.22G/4.38G [00:49<01:50, 28.7MiB/s]
28%|██▊ | 1.22G/4.38G [00:49<02:10, 24.2MiB/s]
28%|██▊ | 1.22G/4.38G [00:49<02:41, 19.5MiB/s]
28%|██▊ | 1.23G/4.38G [00:49<02:09, 24.3MiB/s]
28%|██▊ | 1.23G/4.38G [00:50<02:47, 18.8MiB/s]
28%|██▊ | 1.24G/4.38G [00:50<03:37, 14.4MiB/s]
28%|██▊ | 1.24G/4.38G [00:50<03:49, 13.7MiB/s]
29%|██▊ | 1.25G/4.38G [00:50<02:18, 22.7MiB/s]
29%|██▊ | 1.25G/4.38G [00:50<02:11, 23.8MiB/s]
29%|██▊ | 1.26G/4.38G [00:51<02:17, 22.7MiB/s]
29%|██▊ | 1.26G/4.38G [00:51<02:33, 20.3MiB/s]
29%|██▉ | 1.27G/4.38G [00:51<01:59, 26.1MiB/s]
29%|██▉ | 1.27G/4.38G [00:51<01:24, 36.6MiB/s]
29%|██▉ | 1.28G/4.38G [00:51<01:37, 31.7MiB/s]
29%|██▉ | 1.28G/4.38G [00:51<01:41, 30.4MiB/s]
29%|██▉ | 1.29G/4.38G [00:52<01:44, 29.7MiB/s]
29%|██▉ | 1.29G/4.38G [00:52<01:44, 29.7MiB/s]
30%|██▉ | 1.29G/4.38G [00:52<02:24, 21.4MiB/s]
30%|██▉ | 1.30G/4.38G [00:52<01:45, 29.2MiB/s]
30%|██▉ | 1.31G/4.38G [00:52<01:33, 32.9MiB/s]
30%|██▉ | 1.31G/4.38G [00:52<01:48, 28.4MiB/s]
30%|███ | 1.32G/4.38G [00:53<01:47, 28.5MiB/s]
30%|███ | 1.32G/4.38G [00:53<01:53, 27.0MiB/s]
30%|███ | 1.32G/4.38G [00:53<02:24, 21.2MiB/s]
30%|███ | 1.33G/4.38G [00:53<02:47, 18.3MiB/s]
30%|███ | 1.33G/4.38G [00:53<02:12, 23.0MiB/s]
30%|███ | 1.33G/4.38G [00:54<02:14, 22.6MiB/s]
31%|███ | 1.34G/4.38G [00:54<01:34, 32.1MiB/s]
31%|███ | 1.34G/4.38G [00:54<01:44, 28.9MiB/s]
31%|███ | 1.35G/4.38G [00:54<02:20, 21.6MiB/s]
31%|███ | 1.35G/4.38G [00:54<02:26, 20.7MiB/s]
31%|███ | 1.36G/4.38G [00:54<01:53, 26.7MiB/s]
31%|███ | 1.36G/4.38G [00:55<02:14, 22.4MiB/s]
31%|███ | 1.37G/4.38G [00:55<01:58, 25.4MiB/s]
31%|███ | 1.37G/4.38G [00:55<02:46, 18.1MiB/s]
31%|███▏ | 1.38G/4.38G [00:55<01:52, 26.6MiB/s]
31%|███▏ | 1.38G/4.38G [00:55<02:04, 24.2MiB/s]
32%|███▏ | 1.38G/4.38G [00:56<02:21, 21.2MiB/s]
32%|███▏ | 1.39G/4.38G [00:56<01:53, 26.4MiB/s]
32%|███▏ | 1.39G/4.38G [00:56<02:04, 24.1MiB/s]
32%|███▏ | 1.40G/4.38G [00:56<01:53, 26.3MiB/s]
32%|███▏ | 1.40G/4.38G [00:56<02:28, 20.0MiB/s]
32%|███▏ | 1.41G/4.38G [00:57<02:01, 24.4MiB/s]
32%|███▏ | 1.41G/4.38G [00:57<02:02, 24.2MiB/s]
32%|███▏ | 1.41G/4.38G [00:57<02:01, 24.4MiB/s]
32%|███▏ | 1.42G/4.38G [00:57<01:29, 33.0MiB/s]
33%|███▎ | 1.42G/4.38G [00:57<02:09, 22.8MiB/s]
33%|███▎ | 1.43G/4.38G [00:58<02:28, 19.9MiB/s]
33%|███▎ | 1.43G/4.38G [00:58<01:54, 25.7MiB/s]
33%|███▎ | 1.44G/4.38G [00:58<02:04, 23.6MiB/s]
33%|███▎ | 1.44G/4.38G [00:58<02:13, 22.1MiB/s]
33%|███▎ | 1.44G/4.38G [00:58<02:29, 19.7MiB/s]
33%|███▎ | 1.45G/4.38G [00:58<02:14, 21.7MiB/s]
33%|███▎ | 1.45G/4.38G [00:59<02:39, 18.4MiB/s]
33%|███▎ | 1.46G/4.38G [00:59<02:35, 18.8MiB/s]
33%|███▎ | 1.46G/4.38G [00:59<02:53, 16.9MiB/s]
33%|███▎ | 1.47G/4.38G [00:59<02:08, 22.8MiB/s]
34%|███▎ | 1.47G/4.38G [00:59<02:15, 21.5MiB/s]
34%|███▎ | 1.47G/4.38G [01:00<01:56, 24.9MiB/s]
34%|███▎ | 1.48G/4.38G [01:00<02:06, 23.0MiB/s]
34%|███▍ | 1.48G/4.38G [01:00<01:52, 25.8MiB/s]
34%|███▍ | 1.49G/4.38G [01:00<01:31, 31.5MiB/s]
34%|███▍ | 1.49G/4.38G [01:01<02:24, 20.0MiB/s]
34%|███▍ | 1.50G/4.38G [01:01<02:03, 23.3MiB/s]
34%|███▍ | 1.51G/4.38G [01:01<01:37, 29.4MiB/s]
35%|███▍ | 1.51G/4.38G [01:01<01:42, 28.0MiB/s]
35%|███▍ | 1.52G/4.38G [01:01<01:46, 27.0MiB/s]
35%|███▍ | 1.52G/4.38G [01:01<01:52, 25.4MiB/s]
35%|███▍ | 1.53G/4.38G [01:02<01:44, 27.4MiB/s]
35%|███▍ | 1.53G/4.38G [01:02<02:01, 23.5MiB/s]
35%|███▌ | 1.53G/4.38G [01:02<01:42, 27.7MiB/s]
35%|███▌ | 1.54G/4.38G [01:02<01:46, 26.8MiB/s]
35%|███▌ | 1.54G/4.38G [01:02<01:38, 28.9MiB/s]
35%|███▌ | 1.54G/4.38G [01:02<01:50, 25.8MiB/s]
35%|███▌ | 1.55G/4.38G [01:03<01:42, 27.7MiB/s]
36%|███▌ | 1.56G/4.38G [01:03<01:36, 29.3MiB/s]
36%|███▌ | 1.56G/4.38G [01:03<01:34, 29.8MiB/s]
36%|███▌ | 1.56G/4.38G [01:03<02:08, 22.0MiB/s]
36%|███▌ | 1.57G/4.38G [01:03<02:03, 22.9MiB/s]
36%|███▌ | 1.57G/4.38G [01:03<02:28, 19.0MiB/s]
36%|███▌ | 1.58G/4.38G [01:04<02:00, 23.4MiB/s]
36%|███▌ | 1.58G/4.38G [01:04<02:19, 20.1MiB/s]
36%|███▌ | 1.58G/4.38G [01:04<02:09, 21.6MiB/s]
36%|███▌ | 1.59G/4.38G [01:04<02:25, 19.2MiB/s]
36%|███▋ | 1.59G/4.38G [01:04<01:43, 26.9MiB/s]
36%|███▋ | 1.60G/4.38G [01:04<01:49, 25.4MiB/s]
37%|███▋ | 1.60G/4.38G [01:05<01:40, 27.6MiB/s]
37%|███▋ | 1.60G/4.38G [01:05<01:52, 24.7MiB/s]
37%|███▋ | 1.61G/4.38G [01:05<01:40, 27.7MiB/s]
37%|███▋ | 1.61G/4.38G [01:05<01:52, 24.6MiB/s]
37%|███▋ | 1.62G/4.38G [01:05<01:42, 27.0MiB/s]
37%|███▋ | 1.62G/4.38G [01:05<02:01, 22.7MiB/s]
37%|███▋ | 1.62G/4.38G [01:06<02:04, 22.2MiB/s]
37%|███▋ | 1.62G/4.38G [01:06<02:15, 20.3MiB/s]
37%|███▋ | 1.63G/4.38G [01:06<02:44, 16.8MiB/s]
37%|███▋ | 1.63G/4.38G [01:06<02:43, 16.9MiB/s]
37%|███▋ | 1.64G/4.38G [01:06<01:47, 25.6MiB/s]
37%|███▋ | 1.64G/4.38G [01:06<01:43, 26.5MiB/s]
37%|███▋ | 1.64G/4.38G [01:06<01:42, 26.7MiB/s]
38%|███▊ | 1.65G/4.38G [01:07<02:22, 19.1MiB/s]
38%|███▊ | 1.65G/4.38G [01:07<02:46, 16.4MiB/s]
38%|███▊ | 1.65G/4.38G [01:07<02:54, 15.6MiB/s]
38%|███▊ | 1.66G/4.38G [01:07<01:42, 26.5MiB/s]
38%|███▊ | 1.66G/4.38G [01:08<01:50, 24.6MiB/s]
38%|███▊ | 1.67G/4.38G [01:08<01:34, 28.8MiB/s]
38%|███▊ | 1.67G/4.38G [01:08<01:37, 27.9MiB/s]
38%|███▊ | 1.68G/4.38G [01:08<01:56, 23.3MiB/s]
38%|███▊ | 1.68G/4.38G [01:08<02:04, 21.7MiB/s]
38%|███▊ | 1.68G/4.38G [01:08<01:41, 26.6MiB/s]
39%|███▊ | 1.69G/4.38G [01:09<02:02, 22.0MiB/s]
39%|███▊ | 1.70G/4.38G [01:09<01:21, 32.8MiB/s]
39%|███▉ | 1.70G/4.38G [01:09<01:35, 28.1MiB/s]
39%|███▉ | 1.70G/4.38G [01:09<01:41, 26.4MiB/s]
39%|███▉ | 1.71G/4.38G [01:09<01:21, 32.8MiB/s]
39%|███▉ | 1.71G/4.38G [01:09<01:24, 31.7MiB/s]
39%|███▉ | 1.72G/4.38G [01:09<01:37, 27.3MiB/s]
39%|███▉ | 1.72G/4.38G [01:10<01:59, 22.2MiB/s]
39%|███▉ | 1.73G/4.38G [01:10<01:46, 25.0MiB/s]
39%|███▉ | 1.73G/4.38G [01:10<02:03, 21.6MiB/s]
40%|███▉ | 1.74G/4.38G [01:10<01:44, 25.2MiB/s]
40%|███▉ | 1.74G/4.38G [01:10<01:14, 35.2MiB/s]
40%|███▉ | 1.75G/4.38G [01:11<01:24, 31.0MiB/s]
40%|████ | 1.75G/4.38G [01:11<01:27, 29.9MiB/s]
40%|████ | 1.76G/4.38G [01:11<01:16, 34.2MiB/s]
40%|████ | 1.76G/4.38G [01:11<01:33, 28.1MiB/s]
40%|████ | 1.77G/4.38G [01:11<01:31, 28.6MiB/s]
40%|████ | 1.77G/4.38G [01:12<02:06, 20.7MiB/s]
41%|████ | 1.78G/4.38G [01:12<02:21, 18.5MiB/s]
41%|████ | 1.78G/4.38G [01:12<02:27, 17.6MiB/s]
41%|████ | 1.79G/4.38G [01:12<01:59, 21.8MiB/s]
41%|████ | 1.79G/4.38G [01:12<02:19, 18.6MiB/s]
41%|████ | 1.79G/4.38G [01:13<01:57, 22.1MiB/s]
41%|████ | 1.80G/4.38G [01:13<02:00, 21.5MiB/s]
41%|████ | 1.80G/4.38G [01:13<01:35, 27.1MiB/s]
41%|████ | 1.80G/4.38G [01:13<01:45, 24.3MiB/s]
41%|████▏ | 1.81G/4.38G [01:13<01:29, 28.6MiB/s]
41%|████▏ | 1.81G/4.38G [01:13<01:34, 27.1MiB/s]
42%|████▏ | 1.82G/4.38G [01:14<01:31, 28.0MiB/s]
42%|████▏ | 1.82G/4.38G [01:14<01:31, 27.8MiB/s]
42%|████▏ | 1.83G/4.38G [01:14<01:53, 22.5MiB/s]
42%|████▏ | 1.84G/4.38G [01:14<01:24, 30.2MiB/s]
42%|████▏ | 1.84G/4.38G [01:14<01:39, 25.5MiB/s]
42%|████▏ | 1.85G/4.38G [01:15<01:36, 26.2MiB/s]
42%|████▏ | 1.85G/4.38G [01:15<01:20, 31.4MiB/s]
42%|████▏ | 1.86G/4.38G [01:15<01:20, 31.3MiB/s]
42%|████▏ | 1.86G/4.38G [01:15<01:21, 30.8MiB/s]
43%|████▎ | 1.86G/4.38G [01:15<01:37, 26.0MiB/s]
43%|████▎ | 1.87G/4.38G [01:15<01:36, 26.0MiB/s]
43%|████▎ | 1.87G/4.38G [01:15<01:42, 24.5MiB/s]
43%|████▎ | 1.88G/4.38G [01:16<01:49, 22.9MiB/s]
43%|████▎ | 1.88G/4.38G [01:16<02:12, 18.9MiB/s]
43%|████▎ | 1.89G/4.38G [01:16<01:35, 26.2MiB/s]
43%|████▎ | 1.89G/4.38G [01:16<01:44, 23.9MiB/s]
43%|████▎ | 1.89G/4.38G [01:17<01:51, 22.2MiB/s]
43%|████▎ | 1.90G/4.38G [01:17<02:03, 20.2MiB/s]
43%|████▎ | 1.90G/4.38G [01:17<01:23, 29.8MiB/s]
44%|████▎ | 1.91G/4.38G [01:17<01:39, 24.9MiB/s]
44%|████▎ | 1.91G/4.38G [01:17<01:37, 25.4MiB/s]
44%|████▎ | 1.91G/4.38G [01:17<01:58, 20.8MiB/s]
44%|████▍ | 1.92G/4.38G [01:17<01:34, 26.1MiB/s]
44%|████▍ | 1.92G/4.38G [01:18<01:42, 23.9MiB/s]
44%|████▍ | 1.93G/4.38G [01:18<01:16, 32.2MiB/s]
44%|████▍ | 1.93G/4.38G [01:18<01:33, 26.1MiB/s]
44%|████▍ | 1.94G/4.38G [01:18<01:38, 24.9MiB/s]
44%|████▍ | 1.94G/4.38G [01:18<01:42, 23.9MiB/s]
44%|████▍ | 1.95G/4.38G [01:18<01:18, 31.2MiB/s]
45%|████▍ | 1.95G/4.38G [01:19<01:19, 30.7MiB/s]
45%|████▍ | 1.96G/4.38G [01:19<01:29, 27.0MiB/s]
45%|████▍ | 1.96G/4.38G [01:19<01:35, 25.3MiB/s]
45%|████▍ | 1.96G/4.38G [01:19<01:39, 24.3MiB/s]
45%|████▍ | 1.97G/4.38G [01:19<01:30, 26.8MiB/s]
45%|████▍ | 1.97G/4.38G [01:20<02:07, 18.8MiB/s]
45%|████▌ | 1.97G/4.38G [01:20<02:08, 18.8MiB/s]
45%|████▌ | 1.98G/4.38G [01:20<01:35, 25.0MiB/s]
45%|████▌ | 1.98G/4.38G [01:20<01:49, 22.0MiB/s]
45%|████▌ | 1.99G/4.38G [01:20<01:31, 26.2MiB/s]
45%|████▌ | 1.99G/4.38G [01:20<01:37, 24.7MiB/s]
46%|████▌ | 2.00G/4.38G [01:20<01:16, 31.3MiB/s]
46%|████▌ | 2.00G/4.38G [01:21<01:32, 25.8MiB/s]
46%|████▌ | 2.00G/4.38G [01:21<01:49, 21.7MiB/s]
46%|████▌ | 2.01G/4.38G [01:21<01:32, 25.6MiB/s]
46%|████▌ | 2.01G/4.38G [01:21<01:34, 25.1MiB/s]
46%|████▌ | 2.01G/4.38G [01:21<01:45, 22.4MiB/s]
46%|████▌ | 2.02G/4.38G [01:21<01:49, 21.5MiB/s]
46%|████▌ | 2.02G/4.38G [01:22<02:06, 18.7MiB/s]
46%|████▋ | 2.03G/4.38G [01:22<01:35, 24.7MiB/s]
46%|████▋ | 2.03G/4.38G [01:22<01:38, 23.8MiB/s]
46%|████▋ | 2.04G/4.38G [01:22<01:47, 21.8MiB/s]
47%|████▋ | 2.04G/4.38G [01:22<01:47, 21.7MiB/s]
47%|████▋ | 2.04G/4.38G [01:23<01:50, 21.1MiB/s]
47%|████▋ | 2.04G/4.38G [01:23<01:53, 20.6MiB/s]
47%|████▋ | 2.05G/4.38G [01:23<01:56, 20.1MiB/s]
47%|████▋ | 2.05G/4.38G [01:23<02:05, 18.6MiB/s]
47%|████▋ | 2.05G/4.38G [01:23<01:45, 22.1MiB/s]
47%|████▋ | 2.06G/4.38G [01:23<02:03, 18.9MiB/s]
47%|████▋ | 2.06G/4.38G [01:24<01:46, 21.8MiB/s]
47%|████▋ | 2.06G/4.38G [01:24<02:01, 19.1MiB/s]
47%|████▋ | 2.07G/4.38G [01:24<01:36, 23.9MiB/s]
47%|████▋ | 2.07G/4.38G [01:24<01:35, 24.1MiB/s]
47%|████▋ | 2.08G/4.38G [01:24<01:24, 27.1MiB/s]
48%|████▊ | 2.09G/4.38G [01:25<01:49, 20.9MiB/s]
48%|████▊ | 2.09G/4.38G [01:25<01:50, 20.7MiB/s]
48%|████▊ | 2.10G/4.38G [01:25<01:38, 23.2MiB/s]
48%|████▊ | 2.10G/4.38G [01:25<01:40, 22.7MiB/s]
48%|████▊ | 2.10G/4.38G [01:25<01:23, 27.3MiB/s]
48%|████▊ | 2.11G/4.38G [01:26<01:46, 21.4MiB/s]
48%|████▊ | 2.11G/4.38G [01:26<01:56, 19.4MiB/s]
48%|████▊ | 2.11G/4.38G [01:26<02:01, 18.6MiB/s]
48%|████▊ | 2.12G/4.38G [01:26<01:38, 23.0MiB/s]
48%|████▊ | 2.12G/4.38G [01:26<02:00, 18.7MiB/s]
49%|████▊ | 2.13G/4.38G [01:27<01:26, 26.1MiB/s]
49%|████▉ | 2.14G/4.38G [01:27<01:28, 25.3MiB/s]
49%|████▉ | 2.14G/4.38G [01:27<01:34, 23.6MiB/s]
49%|████▉ | 2.15G/4.38G [01:27<01:30, 24.8MiB/s]
49%|████▉ | 2.15G/4.38G [01:27<01:33, 23.9MiB/s]
49%|████▉ | 2.15G/4.38G [01:28<01:47, 20.7MiB/s]
49%|████▉ | 2.15G/4.38G [01:28<01:33, 23.9MiB/s]
49%|████▉ | 2.16G/4.38G [01:28<02:07, 17.4MiB/s]
49%|████▉ | 2.16G/4.38G [01:28<02:26, 15.1MiB/s]
49%|████▉ | 2.16G/4.38G [01:29<02:41, 13.7MiB/s]
50%|████▉ | 2.17G/4.38G [01:29<01:33, 23.7MiB/s]
50%|████▉ | 2.18G/4.38G [01:29<01:39, 22.2MiB/s]
50%|████▉ | 2.18G/4.38G [01:29<01:36, 22.8MiB/s]
50%|████▉ | 2.18G/4.38G [01:29<02:00, 18.2MiB/s]
50%|████▉ | 2.19G/4.38G [01:29<01:31, 24.0MiB/s]
50%|█████ | 2.19G/4.38G [01:30<01:49, 20.0MiB/s]
50%|█████ | 2.20G/4.38G [01:30<01:46, 20.5MiB/s]
50%|█████ | 2.20G/4.38G [01:30<01:47, 20.3MiB/s]
50%|█████ | 2.20G/4.38G [01:30<01:37, 22.2MiB/s]
50%|█████ | 2.20G/4.38G [01:30<01:52, 19.4MiB/s]
50%|█████ | 2.21G/4.38G [01:30<02:06, 17.1MiB/s]
51%|█████ | 2.21G/4.38G [01:31<01:20, 27.0MiB/s]
51%|█████ | 2.22G/4.38G [01:31<01:35, 22.7MiB/s]
51%|█████ | 2.22G/4.38G [01:31<01:19, 27.3MiB/s]
51%|█████ | 2.22G/4.38G [01:31<01:27, 24.7MiB/s]
51%|█████ | 2.23G/4.38G [01:31<01:41, 21.1MiB/s]
51%|█████ | 2.23G/4.38G [01:31<01:50, 19.4MiB/s]
51%|█████ | 2.24G/4.38G [01:32<01:25, 25.1MiB/s]
51%|█████ | 2.24G/4.38G [01:32<01:44, 20.5MiB/s]
51%|█████▏ | 2.25G/4.38G [01:32<01:26, 24.6MiB/s]
51%|█████▏ | 2.25G/4.38G [01:32<01:34, 22.6MiB/s]
51%|█████▏ | 2.25G/4.38G [01:32<02:05, 16.9MiB/s]
52%|█████▏ | 2.26G/4.38G [01:33<01:49, 19.3MiB/s]
52%|█████▏ | 2.26G/4.38G [01:33<01:28, 24.0MiB/s]
52%|█████▏ | 2.27G/4.38G [01:33<01:39, 21.2MiB/s]
52%|█████▏ | 2.27G/4.38G [01:33<01:41, 20.7MiB/s]
52%|█████▏ | 2.27G/4.38G [01:33<01:47, 19.6MiB/s]
52%|█████▏ | 2.28G/4.38G [01:34<01:22, 25.5MiB/s]
52%|█████▏ | 2.28G/4.38G [01:34<01:38, 21.3MiB/s]
52%|█████▏ | 2.29G/4.38G [01:34<01:35, 22.0MiB/s]
52%|█████▏ | 2.30G/4.38G [01:34<01:31, 22.7MiB/s]
52%|█████▏ | 2.30G/4.38G [01:35<01:50, 18.8MiB/s]
53%|█████▎ | 2.31G/4.38G [01:35<01:53, 18.2MiB/s]
53%|█████▎ | 2.31G/4.38G [01:35<02:06, 16.4MiB/s]
53%|█████▎ | 2.31G/4.38G [01:35<01:31, 22.5MiB/s]
53%|█████▎ | 2.32G/4.38G [01:36<01:59, 17.3MiB/s]
53%|█████▎ | 2.32G/4.38G [01:36<01:29, 22.9MiB/s]
53%|█████▎ | 2.33G/4.38G [01:36<01:14, 27.4MiB/s]
53%|█████▎ | 2.33G/4.38G [01:36<01:56, 17.7MiB/s]
53%|█████▎ | 2.34G/4.38G [01:36<01:32, 22.1MiB/s]
53%|█████▎ | 2.34G/4.38G [01:37<01:33, 21.7MiB/s]
54%|█████▎ | 2.35G/4.38G [01:37<01:26, 23.4MiB/s]
54%|█████▎ | 2.35G/4.38G [01:37<01:37, 20.9MiB/s]
54%|█████▍ | 2.36G/4.38G [01:37<01:08, 29.5MiB/s]
54%|█████▍ | 2.36G/4.38G [01:37<00:59, 34.1MiB/s]
54%|█████▍ | 2.37G/4.38G [01:37<01:08, 29.4MiB/s]
54%|█████▍ | 2.37G/4.38G [01:38<01:08, 29.4MiB/s]
54%|█████▍ | 2.38G/4.38G [01:38<01:09, 28.6MiB/s]
54%|█████▍ | 2.38G/4.38G [01:38<01:21, 24.5MiB/s]
55%|█████▍ | 2.39G/4.38G [01:39<01:38, 20.2MiB/s]
55%|█████▍ | 2.39G/4.38G [01:39<01:44, 19.1MiB/s]
55%|█████▍ | 2.40G/4.38G [01:39<01:49, 18.1MiB/s]
55%|█████▍ | 2.40G/4.38G [01:39<02:10, 15.2MiB/s]
55%|█████▍ | 2.41G/4.38G [01:40<01:58, 16.7MiB/s]
55%|█████▍ | 2.41G/4.38G [01:40<02:05, 15.7MiB/s]
55%|█████▌ | 2.41G/4.38G [01:40<01:30, 21.7MiB/s]
55%|█████▌ | 2.42G/4.38G [01:40<01:34, 20.8MiB/s]
55%|█████▌ | 2.42G/4.38G [01:40<01:34, 20.9MiB/s]
55%|█████▌ | 2.42G/4.38G [01:40<01:29, 21.9MiB/s]
55%|█████▌ | 2.43G/4.38G [01:40<01:18, 24.8MiB/s]
56%|█████▌ | 2.43G/4.38G [01:41<01:27, 22.4MiB/s]
56%|█████▌ | 2.44G/4.38G [01:41<01:25, 22.6MiB/s]
56%|█████▌ | 2.44G/4.38G [01:41<01:41, 19.1MiB/s]
56%|█████▌ | 2.45G/4.38G [01:41<01:24, 22.8MiB/s]
56%|█████▌ | 2.46G/4.38G [01:42<01:00, 31.9MiB/s]
56%|█████▌ | 2.46G/4.38G [01:42<00:55, 34.5MiB/s]
56%|█████▋ | 2.47G/4.38G [01:42<01:21, 23.4MiB/s]
56%|█████▋ | 2.47G/4.38G [01:42<00:58, 32.7MiB/s]
57%|█████▋ | 2.48G/4.38G [01:42<01:08, 27.9MiB/s]
57%|█████▋ | 2.48G/4.38G [01:43<01:19, 23.9MiB/s]
57%|█████▋ | 2.49G/4.38G [01:43<01:18, 24.0MiB/s]
57%|█████▋ | 2.49G/4.38G [01:43<01:13, 25.9MiB/s]
57%|█████▋ | 2.49G/4.38G [01:43<01:29, 21.0MiB/s]
57%|█████▋ | 2.50G/4.38G [01:43<01:31, 20.5MiB/s]
57%|█████▋ | 2.50G/4.38G [01:43<01:30, 20.7MiB/s]
57%|█████▋ | 2.51G/4.38G [01:44<01:08, 27.4MiB/s]
57%|█████▋ | 2.52G/4.38G [01:44<01:08, 27.1MiB/s]
57%|█████▋ | 2.52G/4.38G [01:44<01:17, 24.1MiB/s]
58%|█████▊ | 2.52G/4.38G [01:44<01:14, 24.9MiB/s]
58%|█████▊ | 2.53G/4.38G [01:44<00:54, 34.2MiB/s]
58%|█████▊ | 2.54G/4.38G [01:45<01:10, 26.2MiB/s]
58%|█████▊ | 2.54G/4.38G [01:45<01:09, 26.3MiB/s]
58%|█████▊ | 2.54G/4.38G [01:45<01:15, 24.4MiB/s]
58%|█████▊ | 2.55G/4.38G [01:45<01:08, 26.7MiB/s]
58%|█████▊ | 2.55G/4.38G [01:45<01:11, 25.5MiB/s]
58%|█████▊ | 2.55G/4.38G [01:46<01:28, 20.8MiB/s]
58%|█████▊ | 2.56G/4.38G [01:46<02:23, 12.7MiB/s]
58%|█████▊ | 2.56G/4.38G [01:46<02:27, 12.3MiB/s]
59%|█████▊ | 2.57G/4.38G [01:47<02:02, 14.9MiB/s]
59%|█████▊ | 2.57G/4.38G [01:47<02:20, 12.9MiB/s]
59%|█████▊ | 2.57G/4.38G [01:47<01:49, 16.5MiB/s]
59%|█████▉ | 2.58G/4.38G [01:47<02:01, 14.9MiB/s]
59%|█████▉ | 2.58G/4.38G [01:47<01:33, 19.2MiB/s]
59%|█████▉ | 2.58G/4.38G [01:48<01:48, 16.5MiB/s]
59%|█████▉ | 2.59G/4.38G [01:48<01:39, 18.0MiB/s]
59%|█████▉ | 2.59G/4.38G [01:48<01:33, 19.1MiB/s]
59%|█████▉ | 2.60G/4.38G [01:48<01:33, 19.0MiB/s]
59%|█████▉ | 2.60G/4.38G [01:48<01:34, 18.9MiB/s]
60%|█████▉ | 2.61G/4.38G [01:49<01:10, 25.0MiB/s]
60%|█████▉ | 2.61G/4.38G [01:49<01:16, 23.1MiB/s]
60%|█████▉ | 2.62G/4.38G [01:49<01:07, 26.2MiB/s]
60%|█████▉ | 2.62G/4.38G [01:49<01:05, 26.7MiB/s]
60%|█████▉ | 2.62G/4.38G [01:49<00:59, 29.6MiB/s]
60%|█████▉ | 2.63G/4.38G [01:49<01:03, 27.7MiB/s]
60%|██████ | 2.63G/4.38G [01:49<01:01, 28.4MiB/s]
60%|██████ | 2.64G/4.38G [01:50<01:16, 22.8MiB/s]
60%|██████ | 2.64G/4.38G [01:50<01:04, 27.0MiB/s]
60%|██████ | 2.64G/4.38G [01:50<01:20, 21.6MiB/s]
60%|██████ | 2.65G/4.38G [01:50<01:02, 27.7MiB/s]
61%|██████ | 2.65G/4.38G [01:50<01:11, 24.2MiB/s]
61%|██████ | 2.66G/4.38G [01:50<01:04, 26.8MiB/s]
61%|██████ | 2.66G/4.38G [01:51<01:18, 22.0MiB/s]
61%|██████ | 2.67G/4.38G [01:51<01:19, 21.7MiB/s]
61%|██████ | 2.67G/4.38G [01:51<01:23, 20.5MiB/s]
61%|██████ | 2.67G/4.38G [01:51<01:08, 24.8MiB/s]
61%|██████ | 2.68G/4.38G [01:51<01:16, 22.4MiB/s]
61%|██████ | 2.68G/4.38G [01:52<01:41, 16.7MiB/s]
61%|██████▏ | 2.69G/4.38G [01:52<01:32, 18.3MiB/s]
61%|██████▏ | 2.69G/4.38G [01:52<01:18, 21.4MiB/s]
61%|██████▏ | 2.69G/4.38G [01:52<01:20, 20.9MiB/s]
62%|██████▏ | 2.70G/4.38G [01:53<01:08, 24.6MiB/s]
62%|██████▏ | 2.70G/4.38G [01:53<01:12, 23.0MiB/s]
62%|██████▏ | 2.71G/4.38G [01:53<00:50, 33.1MiB/s]
62%|██████▏ | 2.72G/4.38G [01:53<00:39, 42.6MiB/s]
62%|██████▏ | 2.72G/4.38G [01:53<00:44, 37.3MiB/s]
62%|██████▏ | 2.73G/4.38G [01:53<01:05, 25.2MiB/s]
62%|██████▏ | 2.73G/4.38G [01:53<00:58, 28.0MiB/s]
62%|██████▏ | 2.73G/4.38G [01:54<01:07, 24.5MiB/s]
63%|██████▎ | 2.74G/4.38G [01:54<00:54, 30.2MiB/s]
63%|██████▎ | 2.74G/4.38G [01:54<01:11, 23.0MiB/s]
63%|██████▎ | 2.75G/4.38G [01:55<01:42, 15.9MiB/s]
63%|██████▎ | 2.75G/4.38G [01:55<01:44, 15.5MiB/s]
63%|██████▎ | 2.75G/4.38G [01:55<01:47, 15.2MiB/s]
63%|██████▎ | 2.76G/4.38G [01:55<01:29, 18.1MiB/s]
63%|██████▎ | 2.76G/4.38G [01:55<02:07, 12.7MiB/s]
63%|██████▎ | 2.76G/4.38G [01:56<02:09, 12.5MiB/s]
63%|██████▎ | 2.76G/4.38G [01:56<02:10, 12.4MiB/s]
63%|██████▎ | 2.77G/4.38G [01:56<01:36, 16.7MiB/s]
63%|██████▎ | 2.77G/4.38G [01:56<01:49, 14.7MiB/s]
63%|██████▎ | 2.78G/4.38G [01:56<01:22, 19.4MiB/s]
64%|██████▎ | 2.78G/4.38G [01:57<01:08, 23.4MiB/s]
64%|██████▎ | 2.79G/4.38G [01:57<01:17, 20.7MiB/s]
64%|██████▎ | 2.79G/4.38G [01:57<01:04, 24.7MiB/s]
64%|██████▍ | 2.79G/4.38G [01:57<01:10, 22.4MiB/s]
64%|██████▍ | 2.80G/4.38G [01:57<01:01, 25.9MiB/s]
64%|██████▍ | 2.80G/4.38G [01:57<01:08, 23.0MiB/s]
64%|██████▍ | 2.81G/4.38G [01:58<00:50, 31.3MiB/s]
64%|██████▍ | 2.82G/4.38G [01:58<00:43, 36.1MiB/s]
64%|██████▍ | 2.82G/4.38G [01:58<00:59, 26.0MiB/s]
64%|██████▍ | 2.83G/4.38G [01:58<00:58, 26.5MiB/s]
65%|██████▍ | 2.83G/4.38G [01:58<01:07, 22.9MiB/s]
65%|██████▍ | 2.83G/4.38G [01:59<01:11, 21.8MiB/s]
65%|██████▍ | 2.84G/4.38G [01:59<01:34, 16.4MiB/s]
65%|██████▍ | 2.84G/4.38G [01:59<01:07, 22.9MiB/s]
65%|██████▍ | 2.85G/4.38G [01:59<01:17, 19.8MiB/s]
65%|██████▌ | 2.85G/4.38G [01:59<01:03, 24.1MiB/s]
65%|██████▌ | 2.85G/4.38G [01:59<01:06, 23.0MiB/s]
65%|██████▌ | 2.86G/4.38G [02:00<00:47, 32.1MiB/s]
65%|██████▌ | 2.86G/4.38G [02:00<00:47, 31.8MiB/s]
65%|██████▌ | 2.87G/4.38G [02:00<00:52, 28.9MiB/s]
66%|██████▌ | 2.88G/4.38G [02:00<00:46, 32.5MiB/s]
66%|██████▌ | 2.88G/4.38G [02:00<00:49, 30.2MiB/s]
66%|██████▌ | 2.89G/4.38G [02:00<00:49, 30.5MiB/s]
66%|██████▌ | 2.89G/4.38G [02:01<00:36, 40.5MiB/s]
66%|██████▌ | 2.90G/4.38G [02:01<00:38, 38.5MiB/s]
66%|██████▋ | 2.90G/4.38G [02:01<00:43, 33.9MiB/s]
66%|██████▋ | 2.91G/4.38G [02:01<00:57, 25.4MiB/s]
66%|██████▋ | 2.91G/4.38G [02:01<01:01, 24.0MiB/s]
67%|██████▋ | 2.92G/4.38G [02:02<01:18, 18.7MiB/s]
67%|██████▋ | 2.92G/4.38G [02:02<01:21, 18.0MiB/s]
67%|██████▋ | 2.93G/4.38G [02:02<01:05, 22.1MiB/s]
67%|██████▋ | 2.93G/4.38G [02:02<01:15, 19.3MiB/s]
67%|██████▋ | 2.93G/4.38G [02:03<01:35, 15.2MiB/s]
67%|██████▋ | 2.93G/4.38G [02:03<01:38, 14.8MiB/s]
67%|██████▋ | 2.94G/4.38G [02:03<01:39, 14.6MiB/s]
67%|██████▋ | 2.94G/4.38G [02:03<01:28, 16.2MiB/s]
67%|██████▋ | 2.94G/4.38G [02:04<01:40, 14.3MiB/s]
67%|██████▋ | 2.95G/4.38G [02:04<01:40, 14.3MiB/s]
67%|██████▋ | 2.95G/4.38G [02:04<01:16, 18.6MiB/s]
67%|██████▋ | 2.95G/4.38G [02:04<01:24, 16.9MiB/s]
68%|██████▊ | 2.96G/4.38G [02:04<01:17, 18.4MiB/s]
68%|██████▊ | 2.96G/4.38G [02:05<01:31, 15.5MiB/s]
68%|██████▊ | 2.97G/4.38G [02:05<01:13, 19.2MiB/s]
68%|██████▊ | 2.97G/4.38G [02:05<01:19, 17.7MiB/s]
68%|██████▊ | 2.98G/4.38G [02:05<00:56, 25.0MiB/s]
68%|██████▊ | 2.98G/4.38G [02:05<01:17, 18.1MiB/s]
68%|██████▊ | 2.99G/4.38G [02:06<01:03, 22.1MiB/s]
68%|██████▊ | 2.99G/4.38G [02:06<01:05, 21.3MiB/s]
68%|██████▊ | 3.00G/4.38G [02:06<01:19, 17.4MiB/s]
69%|██████▊ | 3.00G/4.38G [02:06<01:05, 21.2MiB/s]
69%|██████▊ | 3.00G/4.38G [02:07<01:17, 17.7MiB/s]
69%|██████▊ | 3.01G/4.38G [02:07<00:54, 25.3MiB/s]
69%|██████▉ | 3.02G/4.38G [02:07<01:00, 22.5MiB/s]
69%|██████▉ | 3.02G/4.38G [02:07<01:14, 18.3MiB/s]
69%|██████▉ | 3.03G/4.38G [02:08<00:57, 23.4MiB/s]
69%|██████▉ | 3.03G/4.38G [02:08<00:52, 25.7MiB/s]
69%|██████▉ | 3.04G/4.38G [02:08<00:59, 22.7MiB/s]
69%|██████▉ | 3.04G/4.38G [02:08<00:50, 26.7MiB/s]
69%|██████▉ | 3.04G/4.38G [02:08<00:53, 24.8MiB/s]
70%|██████▉ | 3.05G/4.38G [02:08<00:59, 22.4MiB/s]
70%|██████▉ | 3.05G/4.38G [02:09<00:51, 26.0MiB/s]
70%|██████▉ | 3.06G/4.38G [02:09<00:39, 33.4MiB/s]
70%|██████▉ | 3.06G/4.38G [02:09<00:50, 25.8MiB/s]
70%|███████ | 3.07G/4.38G [02:09<00:50, 25.9MiB/s]
70%|███████ | 3.08G/4.38G [02:09<00:47, 27.2MiB/s]
70%|███████ | 3.08G/4.38G [02:10<00:54, 24.0MiB/s]
70%|███████ | 3.08G/4.38G [02:10<00:56, 22.9MiB/s]
70%|███████ | 3.09G/4.38G [02:10<01:16, 16.8MiB/s]
70%|███████ | 3.09G/4.38G [02:10<01:27, 14.8MiB/s]
71%|███████ | 3.10G/4.38G [02:10<01:02, 20.6MiB/s]
71%|███████ | 3.10G/4.38G [02:11<00:44, 28.7MiB/s]
71%|███████ | 3.11G/4.38G [02:11<00:47, 27.0MiB/s]
71%|███████ | 3.11G/4.38G [02:11<00:38, 32.8MiB/s]
71%|███████ | 3.12G/4.38G [02:11<00:44, 28.7MiB/s]
71%|███████ | 3.12G/4.38G [02:11<00:46, 27.1MiB/s]
71%|███████▏ | 3.13G/4.38G [02:11<00:48, 25.8MiB/s]
71%|███████▏ | 3.13G/4.38G [02:12<00:49, 25.4MiB/s]
72%|███████▏ | 3.14G/4.38G [02:12<00:44, 27.9MiB/s]
72%|███████▏ | 3.15G/4.38G [02:12<00:33, 37.3MiB/s]
72%|███████▏ | 3.15G/4.38G [02:12<00:36, 34.1MiB/s]
72%|███████▏ | 3.15G/4.38G [02:12<00:41, 29.7MiB/s]
72%|███████▏ | 3.16G/4.38G [02:13<00:51, 23.7MiB/s]
72%|███████▏ | 3.16G/4.38G [02:13<00:46, 26.1MiB/s]
72%|███████▏ | 3.17G/4.38G [02:13<00:38, 31.4MiB/s]
72%|███████▏ | 3.17G/4.38G [02:13<00:45, 26.8MiB/s]
73%|███████▎ | 3.18G/4.38G [02:13<00:46, 25.8MiB/s]
73%|███████▎ | 3.18G/4.38G [02:13<00:51, 23.2MiB/s]
73%|███████▎ | 3.18G/4.38G [02:14<00:56, 21.3MiB/s]
73%|███████▎ | 3.19G/4.38G [02:14<00:49, 23.9MiB/s]
73%|███████▎ | 3.19G/4.38G [02:14<00:44, 26.8MiB/s]
73%|███████▎ | 3.20G/4.38G [02:14<00:52, 22.7MiB/s]
73%|███████▎ | 3.20G/4.38G [02:14<00:48, 24.3MiB/s]
73%|███████▎ | 3.21G/4.38G [02:15<00:58, 20.2MiB/s]
73%|███████▎ | 3.21G/4.38G [02:15<00:48, 24.3MiB/s]
73%|███████▎ | 3.21G/4.38G [02:15<00:50, 23.0MiB/s]
74%|███████▎ | 3.22G/4.38G [02:15<00:42, 27.5MiB/s]
74%|███████▎ | 3.23G/4.38G [02:15<00:39, 29.1MiB/s]
74%|███████▎ | 3.23G/4.38G [02:15<00:45, 25.5MiB/s]
74%|███████▍ | 3.23G/4.38G [02:16<00:46, 24.5MiB/s]
74%|███████▍ | 3.24G/4.38G [02:16<00:54, 21.1MiB/s]
74%|███████▍ | 3.24G/4.38G [02:16<00:54, 21.1MiB/s]
74%|███████▍ | 3.24G/4.38G [02:16<00:49, 23.0MiB/s]
74%|███████▍ | 3.25G/4.38G [02:16<00:56, 20.0MiB/s]
74%|███████▍ | 3.25G/4.38G [02:16<00:49, 23.0MiB/s]
74%|███████▍ | 3.26G/4.38G [02:17<00:57, 19.7MiB/s]
74%|███████▍ | 3.26G/4.38G [02:17<00:41, 27.0MiB/s]
75%|███████▍ | 3.27G/4.38G [02:17<00:53, 20.9MiB/s]
75%|███████▍ | 3.27G/4.38G [02:18<01:03, 17.5MiB/s]
75%|███████▍ | 3.28G/4.38G [02:18<00:58, 18.8MiB/s]
75%|███████▍ | 3.28G/4.38G [02:18<01:09, 15.9MiB/s]
75%|███████▌ | 3.29G/4.38G [02:18<01:06, 16.4MiB/s]
75%|███████▌ | 3.29G/4.38G [02:19<01:13, 14.8MiB/s]
75%|███████▌ | 3.29G/4.38G [02:19<00:55, 19.7MiB/s]
75%|███████▌ | 3.30G/4.38G [02:19<00:59, 18.1MiB/s]
75%|███████▌ | 3.30G/4.38G [02:19<01:01, 17.6MiB/s]
75%|███████▌ | 3.30G/4.38G [02:19<00:44, 24.0MiB/s]
75%|███████▌ | 3.31G/4.38G [02:19<00:56, 19.1MiB/s]
76%|███████▌ | 3.31G/4.38G [02:20<00:48, 22.0MiB/s]
76%|███████▌ | 3.31G/4.38G [02:20<00:52, 20.4MiB/s]
76%|███████▌ | 3.32G/4.38G [02:20<00:46, 22.9MiB/s]
76%|███████▌ | 3.32G/4.38G [02:20<00:55, 19.2MiB/s]
76%|███████▌ | 3.33G/4.38G [02:20<00:42, 25.0MiB/s]
76%|███████▌ | 3.33G/4.38G [02:20<00:45, 23.2MiB/s]
76%|███████▌ | 3.34G/4.38G [02:21<00:32, 31.8MiB/s]
76%|███████▋ | 3.34G/4.38G [02:21<00:36, 28.6MiB/s]
76%|███████▋ | 3.35G/4.38G [02:21<00:44, 23.1MiB/s]
76%|███████▋ | 3.35G/4.38G [02:21<00:45, 22.6MiB/s]
77%|███████▋ | 3.36G/4.38G [02:21<00:36, 27.9MiB/s]
77%|███████▋ | 3.36G/4.38G [02:21<00:34, 30.0MiB/s]
77%|███████▋ | 3.36G/4.38G [02:22<00:57, 17.7MiB/s]
77%|███████▋ | 3.37G/4.38G [02:22<00:43, 23.4MiB/s]
77%|███████▋ | 3.37G/4.38G [02:22<00:50, 20.0MiB/s]
77%|███████▋ | 3.38G/4.38G [02:22<00:40, 24.9MiB/s]
77%|███████▋ | 3.38G/4.38G [02:23<00:41, 24.1MiB/s]
77%|███████▋ | 3.39G/4.38G [02:23<00:36, 27.5MiB/s]
77%|███████▋ | 3.39G/4.38G [02:23<00:37, 26.1MiB/s]
78%|███████▊ | 3.40G/4.38G [02:23<00:29, 33.5MiB/s]
78%|███████▊ | 3.40G/4.38G [02:23<00:32, 29.8MiB/s]
78%|███████▊ | 3.41G/4.38G [02:23<00:32, 29.7MiB/s]
78%|███████▊ | 3.41G/4.38G [02:23<00:29, 32.6MiB/s]
78%|███████▊ | 3.42G/4.38G [02:24<00:32, 30.1MiB/s]
78%|███████▊ | 3.42G/4.38G [02:24<00:29, 32.3MiB/s]
78%|███████▊ | 3.42G/4.38G [02:24<00:43, 22.2MiB/s]
78%|███████▊ | 3.43G/4.38G [02:24<00:47, 19.8MiB/s]
78%|███████▊ | 3.43G/4.38G [02:24<00:48, 19.4MiB/s]
78%|███████▊ | 3.44G/4.38G [02:25<00:39, 24.0MiB/s]
79%|███████▊ | 3.44G/4.38G [02:25<00:44, 21.3MiB/s]
79%|███████▊ | 3.45G/4.38G [02:25<00:34, 27.3MiB/s]
79%|███████▉ | 3.45G/4.38G [02:25<00:30, 30.2MiB/s]
79%|███████▉ | 3.46G/4.38G [02:25<00:36, 25.5MiB/s]
79%|███████▉ | 3.46G/4.38G [02:26<00:36, 25.3MiB/s]
79%|███████▉ | 3.47G/4.38G [02:26<00:41, 22.3MiB/s]
79%|███████▉ | 3.47G/4.38G [02:26<00:34, 26.1MiB/s]
79%|███████▉ | 3.48G/4.38G [02:26<00:35, 25.6MiB/s]
79%|███████▉ | 3.48G/4.38G [02:26<00:39, 22.8MiB/s]
80%|███████▉ | 3.49G/4.38G [02:27<00:34, 25.7MiB/s]
80%|███████▉ | 3.49G/4.38G [02:27<00:44, 20.1MiB/s]
80%|███████▉ | 3.50G/4.38G [02:27<00:50, 17.6MiB/s]
80%|███████▉ | 3.50G/4.38G [02:27<00:54, 16.3MiB/s]
80%|████████ | 3.51G/4.38G [02:28<00:34, 25.6MiB/s]
80%|████████ | 3.51G/4.38G [02:28<00:33, 25.8MiB/s]
80%|████████ | 3.51G/4.38G [02:28<00:33, 25.8MiB/s]
80%|████████ | 3.52G/4.38G [02:28<00:41, 20.9MiB/s]
80%|████████ | 3.52G/4.38G [02:28<00:36, 23.4MiB/s]
80%|████████ | 3.52G/4.38G [02:28<00:41, 20.5MiB/s]
81%|████████ | 3.53G/4.38G [02:29<00:37, 22.4MiB/s]
81%|████████ | 3.54G/4.38G [02:29<00:28, 29.6MiB/s]
81%|████████ | 3.54G/4.38G [02:29<00:31, 26.9MiB/s]
81%|████████ | 3.55G/4.38G [02:29<00:27, 29.9MiB/s]
81%|████████ | 3.55G/4.38G [02:29<00:31, 26.1MiB/s]
81%|████████ | 3.56G/4.38G [02:29<00:33, 24.8MiB/s]
81%|████████ | 3.56G/4.38G [02:30<00:36, 22.7MiB/s]
81%|████████▏ | 3.56G/4.38G [02:30<00:33, 24.3MiB/s]
81%|████████▏ | 3.56G/4.38G [02:30<00:33, 24.6MiB/s]
81%|████████▏ | 3.57G/4.38G [02:30<00:38, 20.9MiB/s]
82%|████████▏ | 3.57G/4.38G [02:30<00:29, 27.1MiB/s]
82%|████████▏ | 3.58G/4.38G [02:30<00:24, 32.6MiB/s]
82%|████████▏ | 3.58G/4.38G [02:31<00:29, 27.1MiB/s]
82%|████████▏ | 3.59G/4.38G [02:31<00:23, 34.1MiB/s]
82%|████████▏ | 3.60G/4.38G [02:31<00:24, 32.7MiB/s]
82%|████████▏ | 3.60G/4.38G [02:31<00:24, 31.9MiB/s]
82%|████████▏ | 3.61G/4.38G [02:31<00:26, 29.7MiB/s]
82%|████████▏ | 3.61G/4.38G [02:31<00:29, 26.4MiB/s]
82%|████████▏ | 3.61G/4.38G [02:32<00:28, 26.7MiB/s]
83%|████████▎ | 3.62G/4.38G [02:32<00:31, 24.7MiB/s]
83%|████████▎ | 3.62G/4.38G [02:32<00:28, 26.5MiB/s]
83%|████████▎ | 3.62G/4.38G [02:32<00:29, 25.9MiB/s]
83%|████████▎ | 3.63G/4.38G [02:32<00:21, 34.5MiB/s]
83%|████████▎ | 3.64G/4.38G [02:32<00:24, 30.6MiB/s]
83%|████████▎ | 3.64G/4.38G [02:32<00:25, 28.7MiB/s]
83%|████████▎ | 3.64G/4.38G [02:33<00:29, 25.3MiB/s]
83%|████████▎ | 3.65G/4.38G [02:33<00:27, 26.6MiB/s]
83%|████████▎ | 3.65G/4.38G [02:33<00:30, 24.2MiB/s]
83%|████████▎ | 3.66G/4.38G [02:33<00:25, 27.9MiB/s]
84%|████████▎ | 3.66G/4.38G [02:33<00:26, 27.7MiB/s]
84%|████████▎ | 3.66G/4.38G [02:33<00:22, 32.3MiB/s]
84%|████████▎ | 3.67G/4.38G [02:34<00:29, 24.4MiB/s]
84%|████████▍ | 3.67G/4.38G [02:34<00:25, 27.6MiB/s]
84%|████████▍ | 3.68G/4.38G [02:34<00:26, 26.2MiB/s]
84%|████████▍ | 3.68G/4.38G [02:34<00:21, 32.5MiB/s]
84%|████████▍ | 3.68G/4.38G [02:34<00:26, 26.3MiB/s]
84%|████████▍ | 3.69G/4.38G [02:34<00:22, 30.8MiB/s]
84%|████████▍ | 3.69G/4.38G [02:35<00:30, 22.4MiB/s]
84%|████████▍ | 3.70G/4.38G [02:35<00:25, 26.3MiB/s]
85%|████████▍ | 3.71G/4.38G [02:35<00:20, 32.6MiB/s]
85%|████████▍ | 3.71G/4.38G [02:35<00:22, 29.9MiB/s]
85%|████████▍ | 3.72G/4.38G [02:35<00:19, 34.1MiB/s]
85%|████████▌ | 3.72G/4.38G [02:35<00:14, 44.4MiB/s]
85%|████████▌ | 3.73G/4.38G [02:35<00:18, 35.1MiB/s]
85%|████████▌ | 3.73G/4.38G [02:36<00:22, 28.5MiB/s]
85%|████████▌ | 3.74G/4.38G [02:36<00:23, 27.0MiB/s]
85%|████████▌ | 3.74G/4.38G [02:36<00:22, 28.0MiB/s]
86%|████████▌ | 3.75G/4.38G [02:36<00:21, 30.0MiB/s]
86%|████████▌ | 3.75G/4.38G [02:36<00:22, 27.6MiB/s]
86%|████████▌ | 3.76G/4.38G [02:37<00:21, 28.9MiB/s]
86%|████████▌ | 3.76G/4.38G [02:37<00:25, 24.5MiB/s]
86%|████████▌ | 3.76G/4.38G [02:37<00:23, 26.5MiB/s]
86%|████████▌ | 3.77G/4.38G [02:37<00:24, 24.9MiB/s]
86%|████████▌ | 3.77G/4.38G [02:37<00:21, 28.6MiB/s]
86%|████████▌ | 3.78G/4.38G [02:37<00:23, 26.1MiB/s]
86%|████████▋ | 3.78G/4.38G [02:37<00:18, 32.4MiB/s]
86%|████████▋ | 3.78G/4.38G [02:38<00:19, 30.5MiB/s]
87%|████████▋ | 3.79G/4.38G [02:38<00:15, 38.5MiB/s]
87%|████████▋ | 3.80G/4.38G [02:38<00:15, 36.8MiB/s]
87%|████████▋ | 3.80G/4.38G [02:38<00:19, 30.4MiB/s]
87%|████████▋ | 3.81G/4.38G [02:38<00:13, 41.8MiB/s]
87%|████████▋ | 3.81G/4.38G [02:38<00:15, 36.5MiB/s]
87%|████████▋ | 3.82G/4.38G [02:38<00:15, 35.5MiB/s]
87%|████████▋ | 3.82G/4.38G [02:39<00:16, 33.9MiB/s]
87%|████████▋ | 3.83G/4.38G [02:39<00:18, 29.8MiB/s]
88%|████████▊ | 3.83G/4.38G [02:39<00:17, 31.8MiB/s]
88%|████████▊ | 3.84G/4.38G [02:39<00:12, 41.8MiB/s]
88%|████████▊ | 3.85G/4.38G [02:39<00:13, 40.5MiB/s]
88%|████████▊ | 3.85G/4.38G [02:40<00:22, 23.7MiB/s]
88%|████████▊ | 3.86G/4.38G [02:40<00:18, 27.8MiB/s]
88%|████████▊ | 3.86G/4.38G [02:40<00:18, 27.8MiB/s]
88%|████████▊ | 3.86G/4.38G [02:40<00:20, 25.7MiB/s]
88%|████████▊ | 3.87G/4.38G [02:40<00:23, 21.9MiB/s]
88%|████████▊ | 3.87G/4.38G [02:40<00:22, 23.1MiB/s]
88%|████████▊ | 3.87G/4.38G [02:40<00:20, 25.1MiB/s]
88%|████████▊ | 3.88G/4.38G [02:41<00:28, 17.8MiB/s]
89%|████████▊ | 3.88G/4.38G [02:41<00:26, 18.5MiB/s]
89%|████████▊ | 3.88G/4.38G [02:41<00:27, 18.3MiB/s]
89%|████████▉ | 3.89G/4.38G [02:41<00:20, 24.1MiB/s]
89%|████████▉ | 3.90G/4.38G [02:42<00:16, 28.6MiB/s]
89%|████████▉ | 3.90G/4.38G [02:42<00:19, 24.0MiB/s]
89%|████████▉ | 3.91G/4.38G [02:42<00:22, 21.4MiB/s]
89%|████████▉ | 3.91G/4.38G [02:42<00:22, 21.1MiB/s]
89%|████████▉ | 3.92G/4.38G [02:42<00:14, 31.1MiB/s]
89%|████████▉ | 3.92G/4.38G [02:42<00:15, 30.6MiB/s]
90%|████████▉ | 3.92G/4.38G [02:43<00:15, 30.2MiB/s]
90%|████████▉ | 3.93G/4.38G [02:43<00:17, 26.1MiB/s]
90%|████████▉ | 3.93G/4.38G [02:43<00:15, 28.6MiB/s]
90%|████████▉ | 3.94G/4.38G [02:43<00:11, 39.2MiB/s]
90%|█████████ | 3.95G/4.38G [02:43<00:12, 33.8MiB/s]
90%|█████████ | 3.95G/4.38G [02:43<00:14, 30.3MiB/s]
90%|█████████ | 3.95G/4.38G [02:44<00:17, 23.9MiB/s]
90%|█████████ | 3.96G/4.38G [02:44<00:17, 24.2MiB/s]
91%|█████████ | 3.97G/4.38G [02:44<00:13, 30.8MiB/s]
91%|█████████ | 3.97G/4.38G [02:44<00:13, 30.0MiB/s]
91%|█████████ | 3.98G/4.38G [02:44<00:11, 35.4MiB/s]
91%|█████████ | 3.98G/4.38G [02:44<00:11, 33.9MiB/s]
91%|█████████ | 3.98G/4.38G [02:44<00:11, 33.5MiB/s]
91%|█████████ | 3.99G/4.38G [02:45<00:10, 37.4MiB/s]
91%|█████████ | 4.00G/4.38G [02:45<00:13, 28.2MiB/s]
91%|█████████▏| 4.00G/4.38G [02:45<00:13, 28.7MiB/s]
91%|█████████▏| 4.00G/4.38G [02:45<00:14, 26.7MiB/s]
91%|█████████▏| 4.01G/4.38G [02:45<00:17, 21.8MiB/s]
92%|█████████▏| 4.01G/4.38G [02:46<00:19, 18.6MiB/s]
92%|█████████▏| 4.02G/4.38G [02:46<00:19, 18.7MiB/s]
92%|█████████▏| 4.02G/4.38G [02:46<00:19, 18.8MiB/s]
92%|█████████▏| 4.03G/4.38G [02:46<00:12, 29.3MiB/s]
92%|█████████▏| 4.03G/4.38G [02:46<00:12, 27.1MiB/s]
92%|█████████▏| 4.03G/4.38G [02:47<00:15, 22.0MiB/s]
92%|█████████▏| 4.04G/4.38G [02:47<00:12, 26.6MiB/s]
92%|█████████▏| 4.05G/4.38G [02:47<00:13, 24.1MiB/s]
92%|█████████▏| 4.05G/4.38G [02:47<00:12, 27.4MiB/s]
93%|█████████▎| 4.06G/4.38G [02:47<00:10, 30.5MiB/s]
93%|█████████▎| 4.06G/4.38G [02:48<00:11, 27.2MiB/s]
93%|█████████▎| 4.07G/4.38G [02:48<00:15, 20.4MiB/s]
93%|█████████▎| 4.07G/4.38G [02:48<00:16, 19.2MiB/s]
93%|█████████▎| 4.08G/4.38G [02:48<00:14, 21.7MiB/s]
93%|█████████▎| 4.08G/4.38G [02:48<00:14, 21.5MiB/s]
93%|█████████▎| 4.08G/4.38G [02:49<00:10, 27.5MiB/s]
93%|█████████▎| 4.09G/4.38G [02:49<00:11, 24.6MiB/s]
93%|█████████▎| 4.09G/4.38G [02:49<00:15, 19.0MiB/s]
93%|█████████▎| 4.09G/4.38G [02:49<00:16, 17.1MiB/s]
94%|█████████▎| 4.10G/4.38G [02:50<00:13, 20.4MiB/s]
94%|█████████▎| 4.10G/4.38G [02:50<00:13, 20.1MiB/s]
94%|█████████▍| 4.11G/4.38G [02:50<00:09, 27.7MiB/s]
94%|█████████▍| 4.11G/4.38G [02:50<00:11, 23.0MiB/s]
94%|█████████▍| 4.12G/4.38G [02:50<00:10, 24.5MiB/s]
94%|█████████▍| 4.12G/4.38G [02:50<00:14, 18.6MiB/s]
94%|█████████▍| 4.13G/4.38G [02:51<00:10, 24.6MiB/s]
94%|█████████▍| 4.13G/4.38G [02:51<00:10, 23.1MiB/s]
94%|█████████▍| 4.13G/4.38G [02:51<00:08, 28.9MiB/s]
94%|█████████▍| 4.14G/4.38G [02:51<00:08, 27.2MiB/s]
95%|█████████▍| 4.14G/4.38G [02:51<00:08, 28.5MiB/s]
95%|█████████▍| 4.15G/4.38G [02:51<00:09, 24.4MiB/s]
95%|█████████▍| 4.15G/4.38G [02:52<00:09, 23.8MiB/s]
95%|█████████▍| 4.15G/4.38G [02:52<00:10, 22.1MiB/s]
95%|█████████▍| 4.16G/4.38G [02:52<00:08, 25.6MiB/s]
95%|█████████▍| 4.16G/4.38G [02:52<00:09, 23.2MiB/s]
95%|█████████▌| 4.17G/4.38G [02:52<00:07, 29.3MiB/s]
95%|█████████▌| 4.17G/4.38G [02:52<00:09, 22.6MiB/s]
95%|█████████▌| 4.18G/4.38G [02:53<00:08, 24.7MiB/s]
95%|█████████▌| 4.18G/4.38G [02:53<00:08, 23.4MiB/s]
96%|█████████▌| 4.18G/4.38G [02:53<00:06, 29.4MiB/s]
96%|█████████▌| 4.19G/4.38G [02:53<00:06, 28.6MiB/s]
96%|█████████▌| 4.19G/4.38G [02:53<00:06, 30.8MiB/s]
96%|█████████▌| 4.20G/4.38G [02:53<00:06, 27.3MiB/s]
96%|█████████▌| 4.20G/4.38G [02:53<00:06, 29.7MiB/s]
96%|█████████▌| 4.20G/4.38G [02:54<00:07, 24.3MiB/s]
96%|█████████▌| 4.21G/4.38G [02:54<00:06, 24.7MiB/s]
96%|█████████▌| 4.21G/4.38G [02:54<00:07, 22.9MiB/s]
96%|█████████▋| 4.22G/4.38G [02:54<00:05, 29.4MiB/s]
96%|█████████▋| 4.22G/4.38G [02:54<00:06, 26.5MiB/s]
96%|█████████▋| 4.23G/4.38G [02:54<00:05, 31.0MiB/s]
97%|█████████▋| 4.23G/4.38G [02:55<00:06, 25.0MiB/s]
97%|█████████▋| 4.23G/4.38G [02:55<00:06, 23.0MiB/s]
97%|█████████▋| 4.24G/4.38G [02:55<00:06, 20.9MiB/s]
97%|█████████▋| 4.24G/4.38G [02:55<00:05, 25.5MiB/s]
97%|█████████▋| 4.25G/4.38G [02:55<00:06, 21.4MiB/s]
97%|█████████▋| 4.25G/4.38G [02:56<00:04, 25.8MiB/s]
97%|█████████▋| 4.26G/4.38G [02:56<00:03, 30.8MiB/s]
97%|█████████▋| 4.26G/4.38G [02:56<00:04, 25.3MiB/s]
97%|█████████▋| 4.26G/4.38G [02:56<00:05, 22.0MiB/s]
97%|█████████▋| 4.27G/4.38G [02:56<00:05, 21.9MiB/s]
98%|█████████▊| 4.27G/4.38G [02:56<00:04, 24.1MiB/s]
98%|█████████▊| 4.28G/4.38G [02:57<00:04, 25.3MiB/s]
98%|█████████▊| 4.29G/4.38G [02:57<00:02, 34.7MiB/s]
98%|█████████▊| 4.29G/4.38G [02:57<00:02, 32.0MiB/s]
98%|█████████▊| 4.29G/4.38G [02:57<00:02, 35.0MiB/s]
98%|█████████▊| 4.30G/4.38G [02:57<00:02, 32.8MiB/s]
98%|█████████▊| 4.30G/4.38G [02:57<00:02, 29.0MiB/s]
98%|█████████▊| 4.30G/4.38G [02:57<00:02, 26.3MiB/s]
98%|█████████▊| 4.31G/4.38G [02:58<00:02, 34.0MiB/s]
99%|█████████▊| 4.32G/4.38G [02:58<00:02, 31.2MiB/s]
99%|█████████▊| 4.32G/4.38G [02:58<00:02, 27.2MiB/s]
99%|█████████▉| 4.33G/4.38G [02:58<00:02, 25.4MiB/s]
99%|█████████▉| 4.33G/4.38G [02:58<00:02, 24.3MiB/s]
99%|█████████▉| 4.34G/4.38G [02:59<00:02, 21.8MiB/s]
99%|█████████▉| 4.34G/4.38G [02:59<00:02, 18.4MiB/s]
99%|█████████▉| 4.35G/4.38G [02:59<00:01, 24.5MiB/s]
99%|█████████▉| 4.35G/4.38G [02:59<00:00, 34.7MiB/s]
99%|█████████▉| 4.36G/4.38G [02:59<00:00, 31.0MiB/s]
100%|█████████▉| 4.36G/4.38G [03:00<00:00, 26.9MiB/s]
100%|█████████▉| 4.36G/4.38G [03:00<00:00, 24.7MiB/s]
100%|█████████▉| 4.37G/4.38G [03:00<00:00, 24.6MiB/s]
100%|█████████▉| 4.37G/4.38G [03:00<00:00, 20.9MiB/s]
100%|█████████▉| 4.37G/4.38G [03:00<00:00, 17.9MiB/s]
100%|█████████▉| 4.38G/4.38G [03:00<00:00, 21.2MiB/s]
100%|█████████▉| 4.38G/4.38G [03:01<00:00, 19.6MiB/s]
100%|██████████| 4.38G/4.38G [03:01<00:00, 24.2MiB/s]
Then we will load the csv files.
dataset_path = os.path.join(download_dir, 'flickr30k_processed')
train_data = pd.read_csv(f'{dataset_path}/train.csv', index_col=0)
val_data = pd.read_csv(f'{dataset_path}/val.csv', index_col=0)
test_data = pd.read_csv(f'{dataset_path}/test.csv', index_col=0)
image_col = "image"
text_col = "caption"
We also need to expand the relative image paths to use their absolute local paths.
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])
train_data[image_col] = train_data[image_col].apply(lambda ele: path_expander(ele, base_folder=dataset_path))
val_data[image_col] = val_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))
Take train_data for example, let’s see how the data look like in the dataframe.
train_data.head()
| caption | image | |
|---|---|---|
| 0 | Two young guys with shaggy hair look at their ... | /home/ci/autogluon/docs/tutorials/multimodal/s... |
| 1 | Two young White males are outside near many bu... | /home/ci/autogluon/docs/tutorials/multimodal/s... |
| 2 | Two men in green shirts are standing in a yard | /home/ci/autogluon/docs/tutorials/multimodal/s... |
| 3 | A man in a blue shirt standing in a garden | /home/ci/autogluon/docs/tutorials/multimodal/s... |
| 4 | Two friends enjoy time spent together | /home/ci/autogluon/docs/tutorials/multimodal/s... |
Each row is one image and text pair, implying that they match each other. Since one image corresponds to five captions in the dataset, we copy each image path five times to build the correspondences. We can visualize one image-text pair.
train_data[text_col][0]
'Two young guys with shaggy hair look at their hands while hanging out in the yard'
pil_img = Image(filename=train_data[image_col][0])
display(pil_img)
To perform evaluation or semantic search, we need to extract the unique image and text items from text_data and add one label column in the test_data.
test_image_data = pd.DataFrame({image_col: test_data[image_col].unique().tolist()})
test_text_data = pd.DataFrame({text_col: test_data[text_col].unique().tolist()})
test_data_with_label = test_data.copy()
test_label_col = "relevance"
test_data_with_label[test_label_col] = [1] * len(test_data)
Initialize Predictor¶
To initialize a predictor for image-text matching, we need to set problem_type as image_text_similarity. query and response refer to the two dataframe columns in which two items in one row should match each other. You can set query=text_col and response=image_col, or query=image_col and response=text_col. In image-text matching, query and response are equivalent.
from autogluon.multimodal import MultiModalPredictor
predictor = MultiModalPredictor(
query=text_col,
response=image_col,
problem_type="image_text_similarity",
eval_metric="recall",
)
By initializing the predictor for image_text_similarity, you have loaded the pretrained CLIP backbone openai/clip-vit-base-patch32.
Directly Evaluate on Test Dataset (Zero-shot)¶
You may be interested in getting the pretrained model’s performance on your data. Let’s compute the text-to-image and image-to-text retrieval scores.
txt_to_img_scores = predictor.evaluate(
data=test_data_with_label,
query_data=test_text_data,
response_data=test_image_data,
label=test_label_col,
cutoffs=[1, 5, 10],
)
img_to_txt_scores = predictor.evaluate(
data=test_data_with_label,
query_data=test_image_data,
response_data=test_text_data,
label=test_label_col,
cutoffs=[1, 5, 10],
)
print(f"txt_to_img_scores: {txt_to_img_scores}")
print(f"img_to_txt_scores: {img_to_txt_scores}")
txt_to_img_scores: {'recall@1': 0.58964, 'recall@5': 0.83533, 'recall@10': 0.90156}
img_to_txt_scores: {'recall@1': 0.15525, 'recall@5': 0.571, 'recall@10': 0.7176}
Here we report the recall, which is the eval_metric in initializing the predictor above. One cutoff value means using the top k retrieved items to calculate the score. You may find that the text-to-image recalls are much higher than the image-to-text recalls. This is because each image is paired with five texts. In image-to-text retrieval, the upper bound of recall@1 is 20%, which means that the top-1 text is correct, but there are totally five texts to retrieve.
Finetune Predictor¶
After measuring the pretrained performance, we can finetune the model on our dataset to see whether we can get improvements. For a quick demo, here we set the time limit to 180 seconds.
predictor.fit(
train_data=train_data,
tuning_data=val_data,
time_limit=180,
)
No path specified. Models will be saved in: "AutogluonModels/ag-20241023_221831"
=================== System Info ===================
AutoGluon Version: 1.1.2b20241023
Python Version: 3.10.13
Operating System: Linux
Platform Machine: x86_64
Platform Version: #1 SMP Tue Sep 24 10:00:37 UTC 2024
CPU Count: 8
Pytorch Version: 2.3.1+cu121
CUDA Version: 12.1
Memory Avail: 27.15 GB / 30.95 GB (87.7%)
Disk Space Avail: 173.28 GB / 255.99 GB (67.7%)
===================================================
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-20241023_221831
```
INFO: Seed set to 0
GPU Count: 1
GPU Count to be Used: 1
GPU 0 Name: Tesla T4
GPU 0 Memory: 0.57GB/15.0GB (Used/Total)
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 | CLIPForImageText | 151 M | train
1 | response_model | CLIPForImageText | 151 M | train
2 | validation_metric | CustomHitRate | 0 | train
3 | loss_func | MultiNegativesSoftmaxLoss | 0 | train
------------------------------------------------------------------------
151 M Trainable params
0 Non-trainable params
151 M Total params
605.109 Total estimated model params size (MB)
INFO: Time limit reached. Elapsed time is 0:03:00. Signaling Trainer to stop.
INFO: Epoch 0, global step 342: 'val_recall' reached 0.56095 (best 0.56095), saving model to '/home/ci/autogluon/docs/tutorials/multimodal/semantic_matching/AutogluonModels/ag-20241023_221831/epoch=0-step=342.ckpt' as top 3
Start to fuse 1 checkpoints via the greedy soup algorithm.
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-20241023_221831")
```
If you are not satisfied with the model, try to increase the training time,
adjust the hyperparameters (https://auto.gluon.ai/stable/tutorials/multimodal/advanced_topics/customization.html),
or post issues on GitHub (https://github.com/autogluon/autogluon/issues).
<autogluon.multimodal.predictor.MultiModalPredictor at 0x7f18ed06fb50>
Evaluate the Finetuned Model on the Test Dataset¶
Now Let’s evaluate the finetuned model. Similarly, we also compute the recalls of text-to-image and image-to-text retrievals.
txt_to_img_scores = predictor.evaluate(
data=test_data_with_label,
query_data=test_text_data,
response_data=test_image_data,
label=test_label_col,
cutoffs=[1, 5, 10],
)
img_to_txt_scores = predictor.evaluate(
data=test_data_with_label,
query_data=test_image_data,
response_data=test_text_data,
label=test_label_col,
cutoffs=[1, 5, 10],
)
print(f"txt_to_img_scores: {txt_to_img_scores}")
print(f"img_to_txt_scores: {img_to_txt_scores}")
txt_to_img_scores: {'recall@1': 0.69728, 'recall@5': 0.90476, 'recall@10': 0.95048}
img_to_txt_scores: {'recall@1': 0.16845, 'recall@5': 0.6686, 'recall@10': 0.815}
We can observe large improvements over the zero-shot predictor. This means that finetuning CLIP on our customized data may help achieve better performance.
Predict Whether Image and Text Match¶
Whether finetuned or not, the predictor can predict whether image and text pairs match.
pred = predictor.predict(test_data.head(5))
print(pred)
0 1
1 1
2 1
3 1
4 1
dtype: int64
Predict Matching Probabilities¶
The predictor can also return to you the matching probabilities.
proba = predictor.predict_proba(test_data.head(5))
print(proba)
0 1
0 0.343475 0.656525
1 0.325580 0.674420
2 0.347025 0.652975
3 0.344637 0.655363
4 0.330329 0.669671
The second column is the probability of being a match.
Extract Embeddings¶
Another common user case is to extract image and text embeddings.
image_embeddings = predictor.extract_embedding({image_col: test_image_data[image_col][:5].tolist()})
print(image_embeddings.shape)
(5, 512)
text_embeddings = predictor.extract_embedding({text_col: test_text_data[text_col][:5].tolist()})
print(text_embeddings.shape)
(5, 512)
Semantic Search¶
We also provide an advanced util function to conduct semantic search. First, given one or more texts, we can search semantically similar images from an image database.
from autogluon.multimodal.utils import semantic_search
text_to_image_hits = semantic_search(
matcher=predictor,
query_data=test_text_data.iloc[[3]],
response_data=test_image_data,
top_k=5,
)
Let’s visualize the text query and top-1 image response.
test_text_data.iloc[[3]]
| caption | |
|---|---|
| 3 | A man in an orange hat starring at something |
pil_img = Image(filename=test_image_data[image_col][text_to_image_hits[0][0]['response_id']])
display(pil_img)
Similarly, given one or more images, we can retrieve texts with similar semantic meanings.
image_to_text_hits = semantic_search(
matcher=predictor,
query_data=test_image_data.iloc[[6]],
response_data=test_text_data,
top_k=5,
)
Let’s visualize the image query and top-1 text response.
pil_img = Image(filename=test_image_data[image_col][6])
display(pil_img)
test_text_data[text_col][image_to_text_hits[0][1]['response_id']]
'Several students waiting outside an igloo'
Other Examples¶
You may go to AutoMM Examples to explore other examples about AutoMM.
Customization¶
To learn how to customize AutoMM, please refer to Customize AutoMM.