Skip to content

Latest commit

 

History

History
193 lines (160 loc) · 5.89 KB

DATASET.md

File metadata and controls

193 lines (160 loc) · 5.89 KB
Raw

Dataset

The datasets on diverse tasks could be used.

Person re-identification

Market-1501

Market-1501 is a widely-used person re-identification dataset, which contains 32,668 annotated bounding boxes of 1,501 identities in 6 cameras. The dataset could be downloaded in link.

# unpack
unzip Market-1501-v15.09.zip
# link the dataset
ln -s Market-1501-v15.09.15 data/market1501

# the structure of the folder is as follows
data
└── market1501
     ├── bounding_box_test
     ├── bounding_box_train
     ├── gt_bbox
     ├── gt_query
     ├── query
     └── readme.txt

# create db for train/query/gallery subsets
python3 tools/create_market_db.py

@inproceedings{zheng2015scalable,
  title={Scalable Person Re-identification: A Benchmark},
  author={Zheng, Liang and Shen, Liyue and Tian, Lu and Wang, Shengjin and Wang, Jingdong and Tian, Qi},
  booktitle={Computer Vision, IEEE International Conference on},
  year={2015}
}

Pedestrian Attribute Recognition

RAP v2.0

The Richly Annotated Pedestrian (RAP) v2.0 is a large-scale datasets. 84,928 images are annotated with 72 kinds of attributes. However, 54 binary attributes are selected in our experiments as usual practice. You should obtain license agreement in link and request data from the author.

# unpack
mkdir -p data/RAPv2
unzip RAP_dataset.zip
unzip RAP_annotation.zip
# link the dataset
ln -s RAP_dataset data/RAPv2/
ln -s RAP_annotation data/RAPv2/
python3 tools/create_rapv2_db.py

# the structure of the folder is as follows
data
└── RAPv2
     ├── RAP_dataset
     └── RAP_annotation

# create db for train/val
python3 tools/create_rapv2_db.py

@article{li2018richly,
    title={A Richly Annotated Pedestrian Dataset for Person Retrieval in Real Surveillance Scenarios},
    author={Li, Dangwei and Zhang, Zhang and Chen, Xiaotang and Huang, Kaiqi},
    journal={IEEE Transactions on Image Processing},
    volume={28},
    number={4},
    pages={1575--1590},
    year={2019},
    publisher={IEEE}
}

Pose Estimation

COCO keypoints

COCO keypoints dataset contains annotations on person detection and keypoint. The images and annotations should be placed as follows,

data
└── coco_keypoints
     ├── annotations
     |    ├── person_keypoints_train2017.json
     |    └── person_keypoints_val2017.json
     └── images
          ├── train2017/
          |    ├── 000000000009.jpg
          |    ├── 000000000025.jpg
          |    └── ...
          └── val2017/
               ├── 000000000139.jpg
               ├── 000000000285.jpg
               └── ...

# create db for train2017
python3 tools/create_coco_keypoints_db.py --dataset coco_keypoints --dataset-split train2017

# create db for val2017
python3 tools/create_coco_keypoints_db.py --dataset coco_keypoints --dataset-split val2017

@inproceedings{lin2014microsoft,
  title={Microsoft coco: Common objects in context},
  author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
  booktitle={European conference on computer vision},
  pages={740--755},
  year={2014},
  organization={Springer}
}

Dense Human Pose Estimation (DensePose)

DensePose-COCO

DensePose-COCO builds the mapping from all human pixels of an RGB image to the 3D surface of the human body. Images and annotations are downloaded from COCO dataset.

# enter dataset root
mkdir -p data/coco_densepose

# get_DensePose_COCO.sh
mkdir -p data/coco_densepose/annotations && cd data/coco_densepose/annotations
wget https://dl.fbaipublicfiles.com/densepose/densepose_coco_2014_train.json
wget https://dl.fbaipublicfiles.com/densepose/densepose_coco_2014_valminusminival.json
wget https://dl.fbaipublicfiles.com/densepose/densepose_coco_2014_minival.json
wget https://dl.fbaipublicfiles.com/densepose/densepose_coco_2014_test.json

# get_densepose_uv.sh
mkdir -p data/coco_densepose/UV_data && cd data/coco_densepose/UV_data
wget https://dl.fbaipublicfiles.com/densepose/densepose_uv_data.tar.gz
tar xvf densepose_uv_data.tar.gz
rm densepose_uv_data.tar.gz

# Download eval_data
mkdir -p data/coco_densepose/eval_data && cd data/coco_densepose/eval_data
wget https://dl.fbaipublicfiles.com/densepose/densepose_eval_data.tar.gz
tar xvf densepose_eval_data.tar.gz
rm densepose_eval_data.tar.gz

data
└── coco_densepose
     ├── annotations
     |    ├── densepose_coco_2014_train.json
     |    ├── densepose_coco_2014_test.json
     |    ├── densepose_coco_2014_minival.json
     |    └── densepose_coco_2014_valminusminival.json
     ├── images
     |    ├── train2014/
     |    |    ├── COCO_train2014_000000000009.jpg
     |    |    ├── COCO_train2014_000000000025.jpg
     |    |    └── ...
     |    └── val2014/
     |         ├── COCO_val2014_000000000042.jpg
     |         ├── COCO_val2014_000000000073.jpg
     |         └── ...
     ├── UV_data
     |    ├── UV_Processed.mat
     |    └── UV_symmetry_transforms.mat
     └── eval_data
          ├── Pdist_matrix.mat
          ├── SMPL_subdiv.mat
          └── SMPL_SUBDIV_TRANSFORM.mat

# create db for train
python3 tools/create_coco_densepose_db.py --dataset coco_densepose --dataset-split train

# create db for minival
python3 tools/create_coco_densepose_db.py --dataset coco_densepose --dataset-split minival

@InProceedings{Guler2018DensePose,
title={DensePose: Dense Human Pose Estimation In The Wild},
author={R\{i}za Alp G\"uler, Natalia Neverova, Iasonas Kokkinos},
journal={The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2018}
}