Update README.md
Browse filesadd annotation format
README.md
CHANGED
@@ -4,7 +4,64 @@ license_name: server-side-public-license
|
|
4 |
license_link: https://www.mongodb.com/licensing/server-side-public-license
|
5 |
---
|
6 |
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
Fashionpedia(FP) has the following splits (with number of images); *Train* (45,623), *Validation* (1,158), and *Test* (2,044). \
|
10 |
FashionFail(FF) has the following splits (with number of images); *Train* (1,344), *Validation* (150), and *Test* (1,001).
|
|
|
4 |
license_link: https://www.mongodb.com/licensing/server-side-public-license
|
5 |
---
|
6 |
|
7 |
+
# FashionFail Dataset
|
8 |
+
|
9 |
+
## Annotation format
|
10 |
+
|
11 |
+
We follow the annotation format of the [COCO dataset](https://cocodataset.org/#format-data). The annotations are stored in the [JSON format](http://www.json.org/) and are organized as follows:
|
12 |
+
|
13 |
+
```
|
14 |
+
{
|
15 |
+
"info" : info,
|
16 |
+
"licenses" : [license],
|
17 |
+
"categories" : [category],
|
18 |
+
"images" : [image],
|
19 |
+
"annotations" : [annotation],
|
20 |
+
}
|
21 |
+
|
22 |
+
info{
|
23 |
+
"year" : int,
|
24 |
+
"version" : str,
|
25 |
+
"description" : str,
|
26 |
+
"contributor" : str,
|
27 |
+
"url" : str,
|
28 |
+
"date_created" : datetime,
|
29 |
+
}
|
30 |
+
|
31 |
+
license{
|
32 |
+
"id" : int,
|
33 |
+
"name" : str,
|
34 |
+
"url" : str,
|
35 |
+
}
|
36 |
+
|
37 |
+
category{
|
38 |
+
"id" : int,
|
39 |
+
"name" : str,
|
40 |
+
"supercategory" : str,
|
41 |
+
}
|
42 |
+
|
43 |
+
image{
|
44 |
+
"id" : int,
|
45 |
+
"file_name" : str,
|
46 |
+
"height" : int,
|
47 |
+
"width" : int,
|
48 |
+
"license" : int,
|
49 |
+
"original_url" : str,
|
50 |
+
}
|
51 |
+
|
52 |
+
annotation{
|
53 |
+
"id" : int,
|
54 |
+
"image_id" : int,
|
55 |
+
"category_id" : int,
|
56 |
+
"area" : int,
|
57 |
+
"iscrowd" : int, # always 0 as instances represent a single object
|
58 |
+
"bbox" : [x,y,width,height],
|
59 |
+
"segmentation" : RLE, # RLE: {"size", (height, widht), "counts": str}
|
60 |
+
}
|
61 |
+
```
|
62 |
+
|
63 |
+
|
64 |
+
### Comparison with Fashionpedia
|
65 |
|
66 |
Fashionpedia(FP) has the following splits (with number of images); *Train* (45,623), *Validation* (1,158), and *Test* (2,044). \
|
67 |
FashionFail(FF) has the following splits (with number of images); *Train* (1,344), *Validation* (150), and *Test* (1,001).
|