Pedestrian_detection.md

Pedestrian Detection Datasets

For Pedestrian Detection, we adopt the COCO format for annotation files and filter samples without pedestrian instances. We provide the filtered json files for each dataset in the peddet folder.

---

CrowdHuman

Please download the dataset from the official website: CrowdHuman, and put the json files into the CrowHuman/annotations folder. The folder structure should be like this:

- CrowHuman
    - Images
    - annotations
        - train.json
        - val.json
    - annotation_train.odgt
    - annotation_val.odgt

ECP (EuroCity Persons)

Please download the dataset from the official website: ECP, and put the json files into the ECP folder. The folder structure should be like this:

- ECP
    - day/
    - ECP_remove_no_person_img.json

CityPersons

Please download the dataset from the official website: CityPersons, and put the json files into the CityPersons folder. The folder structure should be like this:

- CityPersons
    - leftImg8bit_trainvaltest/
        - leftImg8bit/
            - train/
            - val/
            - test/
    - CityPersons_remove_no_person_img.json

WiderPerson

Please download the dataset from the official website: WiderPerson, and put the json files into the WiderPerson folder. The folder structure should be like this:

- WiderPerson
    - Annotations/
    - Evaluation/
    - Images/
    - ReadMe.txt
    - WiderPerson_remove_no_person_img.json
    - test.txt
    - train.txt
    - val.txt

COCO

Please download the dataset from the official website: COCO, and put the json files into the COCO folder. The folder structure should be like this:

- COCO
    - train2017/
    - val2017/
    - coco_person_remove_no_person_img.json

WIDER_Pedestrian

Please download the dataset from the official website: WIDER_Pedestrian, and put the json files into the WIDER_Pedestrian folder. The folder structure should be like this:

- WIDER_Pedestrian
    - Images/
    - WIDER_Pedestrian_remove_no_person_img.json