2. Load from Hub and transform¶

Setup¶

In [ ]:

%pip install detection_datasets

%pip install detection_datasets

In [2]:

from detection_datasets import DetectionDataset, available_in_hub

from detection_datasets import DetectionDataset, available_in_hub

Load from Hub¶

First we check which datasets are available in the Hub, using the default "detection-datasets' repository.

In [3]:

available_in_hub()

available_in_hub()

Out[3]:

['fashionpedia']

We can then load the "fashionpedia" dataset:

In [4]:

dd = DetectionDataset().from_hub(dataset_name='fashionpedia')

dd = DetectionDataset().from_hub(dataset_name='fashionpedia')

Downloading:   0%|          | 0.00/1.78k [00:00<?, ?B/s]

WARNING:datasets.builder:Using custom data configuration detection-datasets--fashionpedia-1960081b1c22758d

Downloading and preparing dataset None/None (download: 3.24 GiB, generated: 3.12 GiB, post-processed: Unknown size, total: 6.36 GiB) to /root/.cache/huggingface/datasets/detection-datasets___parquet/detection-datasets--fashionpedia-1960081b1c22758d/0.0.0/2a3b91fbd88a2c90d1dbbb32b460cf621d31bd5b05b934492fdef7d8d6f236ec...

Downloading data files:   0%|          | 0/2 [00:00<?, ?it/s]

Downloading data:   0%|          | 0.00/84.8M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/482M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/480M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/480M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/490M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/488M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/487M [00:00<?, ?B/s]

Downloading data:   0%|          | 0.00/487M [00:00<?, ?B/s]

Extracting data files:   0%|          | 0/2 [00:00<?, ?it/s]

0 tables [00:00, ? tables/s]

0 tables [00:00, ? tables/s]

Dataset parquet downloaded and prepared to /root/.cache/huggingface/datasets/detection-datasets___parquet/detection-datasets--fashionpedia-1960081b1c22758d/0.0.0/2a3b91fbd88a2c90d1dbbb32b460cf621d31bd5b05b934492fdef7d8d6f236ec. Subsequent calls will reuse this data.

  0%|          | 0/2 [00:00<?, ?it/s]

Moving images from memory to disk:   0%|          | 0/1158 [00:00<?, ?ex/s]

Moving images from memory to disk:   0%|          | 0/45623 [00:00<?, ?ex/s]

Analyze the dataset¶

In [5]:

dd.data

dd.data

Out[5]:

	image_path	width	height	split	bbox_id	bbox	category_id	category	area
image_id
23	/root/.detection-datasets/23.jpg	682	1024	train	[150311, 150312, 150313, 150314]	[Bbox id 150311 [445. 910. 505. 983.], Bbox id...	[23, 23, 33, 10]	[shoe, shoe, neckline, dress]	[1422, 843, 373, 56375]
25	/root/.detection-datasets/25.jpg	683	1024	train	[158953, 158954, 158955, 158956, 158957, 15895...	[Bbox id 158953 [182. 220. 472. 647.], Bbox id...	[2, 33, 31, 31, 13, 7, 22, 22, 23, 23]	[sweater, neckline, sleeve, sleeve, glasses, s...	[87267, 1220, 16895, 18541, 1468, 9360, 8629, ...
26	/root/.detection-datasets/26.jpg	1024	683	train	[169196, 169197, 169198, 169199, 169200, 16920...	[Bbox id 169196 [441. 132. 499. 150.], Bbox id...	[13, 29, 28, 32, 32, 31, 31, 0, 31, 31, 18, 4,...	[glasses, lapel, collar, pocket, pocket, sleev...	[587, 2922, 931, 262, 111, 1171, 540, 3981, 44...
27	/root/.detection-datasets/27.jpg	682	1024	train	[167967, 167968, 167969, 167970, 167971, 16797...	[Bbox id 167967 [300. 421. 460. 846.], Bbox id...	[6, 23, 23, 31, 31, 4, 1, 35, 32, 35, 35, 35, ...	[pants, shoe, shoe, sleeve, sleeve, jacket, to...	[44062, 2140, 2633, 9206, 5905, 44791, 12948, ...
28	/root/.detection-datasets/28.jpg	853	1024	train	[168041, 168042, 168043, 168044, 168045, 16804...	[Bbox id 168041 [ 238. 309. 471. 1022.], Bbo...	[10, 32, 35, 31, 4, 29, 33]	[dress, pocket, zipper, sleeve, jacket, lapel,...	[12132, 1548, 755, 43926, 178328, 9316, 136]
...	...	...	...	...	...	...	...	...	...
50525	/root/.detection-datasets/50525.jpg	660	1024	train	[82929, 82930, 82931, 82932, 82933, 82934, 82935]	[Bbox id 82929 [419. 214. 576. 637.], Bbox id ...	[31, 31, 33, 2, 15, 8, 32]	[sleeve, sleeve, neckline, sweater, headband, ...	[39360, 26851, 2431, 262021, 1422, 34694, 294]
50526	/root/.detection-datasets/50526.jpg	682	1024	train	[82936, 82937, 82938, 82939, 82940]	[Bbox id 82936 [ 168. 846. 504. 1022.], Bbox...	[6, 1, 33, 32, 32]	[pants, top, t-shirt, sweatshirt, neckline, po...	[49375, 151458, 4122, 993, 371]
50528	/root/.detection-datasets/50528.jpg	682	1024	train	[177641, 177642, 177643, 177644, 177645, 177646]	[Bbox id 177641 [195. 256. 408. 725.], Bbox id...	[10, 33, 31, 31, 23, 23]	[dress, neckline, sleeve, sleeve, shoe, shoe]	[47419, 267, 12965, 2090, 2914, 1171]
50530	/root/.detection-datasets/50530.jpg	682	1024	train	[82941, 82942, 82943, 82944, 82945]	[Bbox id 82941 [157. 236. 366. 591.], Bbox id ...	[11, 33, 31, 23, 23]	[jumpsuit, neckline, sleeve, shoe, shoe]	[46325, 684, 2113, 1714, 2927]
50531	/root/.detection-datasets/50531.jpg	682	1024	train	[82946, 82947, 82948, 82949, 82950, 82951, 829...	[Bbox id 82946 [308. 900. 337. 976.], Bbox id ...	[23, 23, 10, 19, 34, 31, 31, 39, 33]	[shoe, shoe, dress, belt, buckle, sleeve, slee...	[919, 2813, 86771, 4029, 384, 2542, 1102, 2923...

46781 rows × 9 columns

In [6]:

dd.show(image_id=25)

dd.show(image_id=25)

Out[6]:

In [7]:

dd.categories

dd.categories

Out[7]:

	category
category_id
0	shirt, blouse
1	top, t-shirt, sweatshirt
2	sweater
3	cardigan
4	jacket
5	vest
6	pants
7	shorts
8	skirt
9	coat
10	dress
11	jumpsuit
12	cape
13	glasses
14	hat
15	headband, head covering, hair accessory
16	tie
17	glove
18	watch
19	belt
20	leg warmer
21	tights, stockings
22	sock
23	shoe
24	bag, wallet
25	scarf
26	umbrella
27	hood
28	collar
29	lapel
30	epaulette
31	sleeve
32	pocket
33	neckline
34	buckle
35	zipper
36	applique
37	bead
38	bow
39	flower
40	fringe
41	ribbon
42	rivet
43	ruffle
44	sequin
45	tassel

Transform the dataset¶

Change the categories¶

In this dataset we have very fine_grained labels, so to create an easier task we may want to focus on higher level labels, like "clothing" instead of "pants".

For example, let us assume that we are only interested in detection these 4 classes:

• Clothing
• Shoes
• Bags
• Accessories

We can start by printing a list of the existing category names, and use it to create our mapping dictionnary:

In [8]:

dd.category_names

dd.category_names

Out[8]:

['shirt, blouse',
 'top, t-shirt, sweatshirt',
 'sweater',
 'cardigan',
 'jacket',
 'vest',
 'pants',
 'shorts',
 'skirt',
 'coat',
 'dress',
 'jumpsuit',
 'cape',
 'glasses',
 'hat',
 'headband, head covering, hair accessory',
 'tie',
 'glove',
 'watch',
 'belt',
 'leg warmer',
 'tights, stockings',
 'sock',
 'shoe',
 'bag, wallet',
 'scarf',
 'umbrella',
 'hood',
 'collar',
 'lapel',
 'epaulette',
 'sleeve',
 'pocket',
 'neckline',
 'buckle',
 'zipper',
 'applique',
 'bead',
 'bow',
 'flower',
 'fringe',
 'ribbon',
 'rivet',
 'ruffle',
 'sequin',
 'tassel']

In [9]:

mapping = {
    'shirt, blouse': 'clothing',
    'top, t-shirt, sweatshirt': 'clothing',
    'sweater': 'clothing',
    'cardigan': 'clothing',
    'jacket': 'clothing',
    'vest': 'clothing',
    'pants': 'clothing',
    'shorts': 'clothing',
    'skirt': 'clothing',
    'coat': 'clothing',
    'dress': 'clothing',
    'jumpsuit': 'clothing',
    'cape': 'clothing',
    'glasses': 'accessories',
    'hat': 'accessories',
    'headband, head covering, hair accessory': 'accessories',
    'tie': 'accessories',
    'glove': 'accessories',
    'belt': 'accessories',
    'tights, stockings': 'accessories',
    'sock': 'accessories',
    'shoe': 'shoes',
    'bag, wallet': 'bags',
    'scarf': 'accessories',
}

mapping = { 'shirt, blouse': 'clothing', 'top, t-shirt, sweatshirt': 'clothing', 'sweater': 'clothing', 'cardigan': 'clothing', 'jacket': 'clothing', 'vest': 'clothing', 'pants': 'clothing', 'shorts': 'clothing', 'skirt': 'clothing', 'coat': 'clothing', 'dress': 'clothing', 'jumpsuit': 'clothing', 'cape': 'clothing', 'glasses': 'accessories', 'hat': 'accessories', 'headband, head covering, hair accessory': 'accessories', 'tie': 'accessories', 'glove': 'accessories', 'belt': 'accessories', 'tights, stockings': 'accessories', 'sock': 'accessories', 'shoe': 'shoes', 'bag, wallet': 'bags', 'scarf': 'accessories', }

We will then map the existing classes to these new classes with a simple Python dictionnary.

3 thinks will happen when applying this new mapping:

• The existing categories will be replaced by their mapped category names
• The existing categories that have no equivalent in this mapping will be dropped
• The category_id will be recreated using the alphabetical order of the new categories (here 'Accessories' will have id 0, 'Bags' 1 and so on)

In [10]:

dd.map_categories(mapping=mapping)

dd.map_categories(mapping=mapping)

Out[10]:

<detection_datasets.detection_dataset.DetectionDataset at 0x7f4fd5cf7d90>

In [11]:

dd.categories

dd.categories

Out[11]:

	category
category_id
0	accessories
1	bags
2	clothing
3	shoes

In [12]:

dd.data

dd.data

Out[12]:

	image_path	width	height	split	bbox_id	bbox	category_id	category	area
image_id
23	/root/.detection-datasets/23.jpg	682	1024	train	[150311, 150312, 150314]	[Bbox id 150311 [445. 910. 505. 983.], Bbox id...	[3, 3, 2]	[shoes, shoes, clothing]	[1422, 843, 56375]
25	/root/.detection-datasets/25.jpg	683	1024	train	[158953, 158957, 158958, 158959, 158960, 15896...	[Bbox id 158953 [182. 220. 472. 647.], Bbox id...	[2, 0, 2, 0, 0, 3, 3]	[clothing, accessories, clothing, accessories,...	[87267, 1468, 9360, 8629, 8270, 2717, 3121]
26	/root/.detection-datasets/26.jpg	1024	683	train	[169196, 169203, 169207, 169208, 169209, 169210]	[Bbox id 169196 [441. 132. 499. 150.], Bbox id...	[0, 2, 2, 2, 3, 3]	[accessories, clothing, clothing, clothing, sh...	[587, 3981, 26621, 16954, 2167, 1773]
27	/root/.detection-datasets/27.jpg	682	1024	train	[167967, 167968, 167969, 167972, 167973]	[Bbox id 167967 [300. 421. 460. 846.], Bbox id...	[2, 3, 3, 2, 2]	[clothing, shoes, shoes, clothing, clothing]	[44062, 2140, 2633, 44791, 12948]
28	/root/.detection-datasets/28.jpg	853	1024	train	[168041, 168045]	[Bbox id 168041 [ 238. 309. 471. 1022.], Bbo...	[2, 2]	[clothing, clothing]	[12132, 178328]
...	...	...	...	...	...	...	...	...	...
50525	/root/.detection-datasets/50525.jpg	660	1024	train	[82932, 82933, 82934]	[Bbox id 82932 [ 6. 209. 576. 923.], Bbox id ...	[2, 0, 2]	[clothing, accessories, clothing]	[262021, 1422, 34694]
50526	/root/.detection-datasets/50526.jpg	682	1024	train	[82936, 82937]	[Bbox id 82936 [ 168. 846. 504. 1022.], Bbox...	[2, 2]	[clothing, clothing]	[49375, 151458]
50528	/root/.detection-datasets/50528.jpg	682	1024	train	[177641, 177645, 177646]	[Bbox id 177641 [195. 256. 408. 725.], Bbox id...	[2, 3, 3]	[clothing, shoes, shoes]	[47419, 2914, 1171]
50530	/root/.detection-datasets/50530.jpg	682	1024	train	[82941, 82944, 82945]	[Bbox id 82941 [157. 236. 366. 591.], Bbox id ...	[2, 3, 3]	[clothing, shoes, shoes]	[46325, 1714, 2927]
50531	/root/.detection-datasets/50531.jpg	682	1024	train	[82946, 82947, 82948, 82949]	[Bbox id 82946 [308. 900. 337. 976.], Bbox id ...	[3, 3, 2, 0]	[shoes, shoes, clothing, accessories]	[919, 2813, 86771, 4029]

46780 rows × 9 columns

In [13]:

dd.show(image_id=25)

dd.show(image_id=25)

Out[13]:

Change the splits¶

The fashionpedia comes with the following splits:

In [14]:

dd.splits

dd.splits

Out[14]:

['train', 'val']

In [15]:

dd.split_proportions

dd.split_proportions

Out[15]:

	train	val	test
0	0.975246	0.024754	0.0

In [16]:

dd.split(splits=[0.9, 0.05, 0.05])

dd.split(splits=[0.9, 0.05, 0.05])

Out[16]:

<detection_datasets.detection_dataset.DetectionDataset at 0x7f4fd5cf7d90>

In [17]:

dd.splits

dd.splits

Out[17]:

['train', 'val', 'test']

In [18]:

dd.split_proportions

dd.split_proportions

Out[18]:

	train	val	test
0	0.9	0.05	0.05

Push to new dataset to the Hub¶

In [ ]:

! huggingface-cli login

! huggingface-cli login

In [21]:

dd.to_hub(dataset_name='fashionpedia_4_categories', repo_name='detection-datasets')

dd.to_hub(dataset_name='fashionpedia_4_categories', repo_name='detection-datasets')

WARNING:datasets.dataset_dict:Pushing split train to the Hub.

  0%|          | 0/7 [00:00<?, ?ba/s]

Pushing dataset shards to the dataset hub:   0%|          | 0/7 [00:00<?, ?it/s]

  0%|          | 0/7 [00:00<?, ?ba/s]

  0%|          | 0/7 [00:00<?, ?ba/s]

  0%|          | 0/7 [00:00<?, ?ba/s]

  0%|          | 0/7 [00:00<?, ?ba/s]

  0%|          | 0/7 [00:00<?, ?ba/s]

  0%|          | 0/7 [00:00<?, ?ba/s]

WARNING:datasets.dataset_dict:Pushing split test to the Hub.

  0%|          | 0/3 [00:00<?, ?ba/s]

Pushing dataset shards to the dataset hub:   0%|          | 0/1 [00:00<?, ?it/s]

WARNING:datasets.dataset_dict:Pushing split val to the Hub.

  0%|          | 0/3 [00:00<?, ?ba/s]

Pushing dataset shards to the dataset hub:   0%|          | 0/1 [00:00<?, ?it/s]

Out[21]:

<detection_datasets.detection_dataset.DetectionDataset at 0x7f4fd5cf7d90>