Acne04-v2 dataset

ACNE04 dataset

Acne04-v2_annotations.json is a json format file containing the annotations our experts made for our paper "AcneAI: A new acne severity assessment method using digital images and deep learning" (early accepted for MICCAI2024, link soon). We provide new labels for the images of the open source dataset ACNE04 from the paper Joint Acne Image Grading and Counting via Label Distribution Learning, Wu et al. ICCV 2019 released in 2019. In order to provide a better quality dataset, we removed the low quality images and asked our experts to annotate more acne lesions.

The news version of the dataset contains 1204 images and 32 443 annotations, and corresponds to a subset of the acne04 first version.

The format of the json file is inspired from the COCO annotations format.

• "images" is a list of dictionnaries, each element corresponding to one image. Each dictionnary is composed of several fields : image id , name of the image file_name, and image size information (height and width).
• "annotations" is a list of dictionnaries, each element corresponding to one acne lesion. Basically, for each image, there are several annotations related. The different fields of each dictionnary are : annotation id, image_id which is the id in the of the source image in the "images" corresponding dictionnary, center coordinates and radius of the acne lesion on the image.
• "info" is a dictionnary containing information about this file : the predictions version, a brief description, the release year, and the names of the contributors.

The file draw_acne_from_json.py is one example of how to plot the circles of acne on the images. In the folder examples/, we provide samples of annotated images (in blue our v2 version and in green the original version).

Citation

Please cite our paper: @InProceedings{10.1007/978-3-031-72086-4_7, author="Gazeau, Lea and Nguyen, Hang and Nguyen, Zung and Lebedeva, Mariia and Nguyen, Thanh and To, Tat-Dat and Le Digabel, Jimmy and Filiol, Jerome and Josse, Gwendal and Perlis, Clifford and Wolfe, Jonathan", title="AcneAI: A New Acne Severity Assessment Method Using Digital Images and Deep Learning", booktitle="Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024", year="2024", publisher="Springer Nature Switzerland", address="Cham", pages="68--78", isbn="978-3-031-72086-4" }

Ackownledgement

This research is supported by Torus AI and BelleTorus Corporation.

We also thank Wu et al for the Acne04 data made public.