AiC Dataset

AiC (Attributes in Crowd) is a novel synthetic dataset for people attribute recognition in presence of strong occlusions created by exploiting the highly photorealistic video game Grand Theft Auto V. It features 125,000 samples, all being a unique person, each of which is automatically labeled with information concerning visual attributes, as well as joint locations.

Obtain the Dataset

You can download AiC here. By downloading the dataset you agree on the following statement: "I declare that I will use the AiC Dataset for research and educational purposes only, since I am aware that commercial use is prohibited. I also undertake to purchase a copy of Grand Theft Auto V."

AiC-Dataset Contents

After the data download, your AiC-Dataset directory will contain the following files:

• crops: directory with image samples. For each sample x we have:

  • x.jpg: fully visible sample
  • x_occ.jpg: occluded sample

• annotations.json: annotation file of the whole dataset

• train.json: train split containing the ids used as training set

• test.json: test split containing the ids used as test set

Annotations

The annotation file consists of a list of dictionaries. Each element of the list is a sample of the dataset. Each dictionary is organized as follows:

Key	Description
attributes	list of binary attributes; see 'Attributes' subsection
pose	list of joints; see 'Joins' subsection
id	unique identifier of the sample

IMPORTANT: given the id, the correspondent fully visible image is crops/id.jpg, while the occluded one is crops/id_occ.jpg.

Attributes

The list of binary attributes is ordered as follows:

 0: Female
 1: Age17-30
 2: Age31-45
 3: BodyNormal
 4: BodyThin
 5: BaldHead
 6: LongHair
 7: BlackHair
 8: Hat
 9: Muffler
10: Shirt
11: Sweater
12: Jacket
13: TightHood
14: ShortSleeve
15: LongTrousers
16: Skirt
17: Jeans
18: Tights
19: shoes-Leather
20: shoes-Sport
21: shoes-Boots
22: Backpack
23: Eyeglasses

Joins

Each joint is a list containing:

Element index	Name	Description
0	joint type	identifier of the type of joint; see 'Joint Types' subsection
1	x2D	2D x coordinate of the joint in pixel
2	y2D	2D y coordinate of the joint in pixel
3	occluded	1 if the joint is occluded; 0 otherwise
4	self-occluded	1 if the joint is occluded by its owner; 0 otherwise

The association between numerical identifier and type of joint is the following:

 0: head_top
 1: head_center
 2: neck
 3: right_clavicle
 4: right_shoulder
 5: right_elbow
 6: right_wrist
 7: left_clavicle
 8: left_shoulder
 9: left_elbow
10: left_wrist
11: spine0
12: spine1
13: spine2
14: spine3
15: spine4
16: right_hip
17: right_knee
18: right_ankle
19: left_hip
20: left_knee
21: left_ankle

Citation

This dataset was introduced in the paper "Can Adversarial Networks Hallucinate Occluded People With a Plausible Aspect?". We believe in open research and we are happy if you find this data useful.
If you use it, please cite our works.

@article{fulgeri2019can,
  title     = {Can Adversarial Networks Hallucinate Occluded People With a Plausible Aspect?},
  author    = {Fulgeri, Federico and Fabbri, Matteo and Alletto, Stefano and Calderara, Simone and Cucchiara, Rita},
  journal   = {arXiv preprint arXiv:1901.08097},
  year      = {2019}
}

@inproceedings{fabbri2018learning,
  title     = {Learning to Detect and Track Visible and Occluded Body Joints in a Virtual World},
  author    = {Fabbri, Matteo and Lanzi, Fabio and Calderara, Simone and Palazzi, Andrea and Vezzani, Roberto and Cucchiara, Rita},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2018}
}