CrowdHuman

This repo contains a script to convert the CrowdHuman dataset annotations to COCO format and a dataset Class for reading data.

Introduction

CrowdHuman is a benchmark dataset to better evaluate detectors in crowd scenarios. The CrowdHuman dataset is large, rich-annotated and contains high diversity. CrowdHuman contains 15000, 4370 and 5000 images for training, validation, and testing, respectively. There are a total of 470K human instances from train and validation subsets and 23 persons per image, with various kinds of occlusions in the dataset. Each human instance is annotated with a head bounding-box, human visible-region bounding-box and human full-body bounding-box. We hope our dataset will serve as a solid baseline and help promote future research in human detection tasks.

Dataset

Annotation format

Supported annotation_train.odgt and annotation_val.odgt which contains the annotations of CrowdHuman.

odgt is a file format that each line of it is a JSON, this JSON contains the whole annotations for the relative image. We prefer using this format since it is reader-friendly.

Annotation format:

JSON{
    "ID" : image_filename,
    "gtboxes" : [gtbox], 
}

gtbox{
    "tag" : "person" or "mask", 
    "vbox": [x, y, w, h],
    "fbox": [x, y, w, h],
    "hbox": [x, y, w, h],
    "extra" : extra, 
    "head_attr" : head_attr, 
}

extra{
    "ignore": 0 or 1,
    "box_id": int,
    "occ": int,
}

head_attr{
    "ignore": 0 or 1,
    "unsure": int,
    "occ": int,
}

• Keys in extra and head_attr are optional, it means some of them may not exist
• extra/head_attr contains attributes for person/head
• tag is mask means that this box is crowd/reflection/something like person/... and need to be ignore(the ignore in extra is 1)
• vbox, fbox, hbox means visible box, full box, head box respectively

Download

CrowdHuman_train01.zip

CrowdHuman_train02.zip

CrowdHuman_train03.zip

CrowdHuman_val.zip

annotation_train.odgt

annotation_val.odgt

CrowdHuman_test.zip

Get Started

Convert annotations

Before converting, ensure that the dataset folder format like this:

|- crowdhuman
 |- Images #contain all train and test images
 |- annotation_train.json
 |- annotation_val.json

You can use this to simply keep full boxes with tag person only

python crowdhuman2coco.py -d /path/to/crowdhuman/dataset -o /path/to/annotation_train.odgt/ -j /path/to/annotation_train.json

For more demand, run this to get more detial infomation

python crowdhuman2coco.py --help

Simple Dataset

This repo also contains two simple implement of CrowdHuman Dataset Class in PyTorch and MegEngine.

The Dataset will return a tuple that contains the annotations that you need in order everytime when it call __getitem__

supported_order

class CrowdHuman(VisionDataset):
    supported_order = (
        "image",
        "boxes",
        "vboxes",
        "hboxes",
        "boxes_category",
        "info",
    )

You can easily use this to instantiate a crowdhuman_dataset

crowdhuman_dataset = CrowdHuman(
    root='path/to/CrowdHuman',
    ann_file='path/to/annotations.json',
    remove_images_without_annotations=True,
    order=[
        'image',
        'boxes',
        'boxes_category'
        'info'
    ]
)