Baselines and a datasheet for the Cerema AWP dataset
Ismaïla Seck Université Clermont Auvergne, LIMOS INSA de Rouen, LITIS Khouloud Dahmane Cerema, Clermont-Ferrand Pierre Duthon Cerema, Clermont-Ferrand Gaëlle Loosli Université Clermont Auvergne, LIMOS
Abstract
This paper presents the recently published Cerema AWP (Adverse Weather Pedestrian) dataset for various machine learning tasks and its exports in machine learning friendly format. We explain why this dataset can be interesting (mainly because it is a greatly controlled and fully annotated image dataset) and present baseline results for various tasks. Moreover, we decided to follow the very recent suggestions of datasheets for dataset, trying to standardize all the available information of the dataset, with a transparency objective.
中文速览
一个在人工智能领域"刷榜"横行的年代，缺乏公平基准（baseline）和数据集透明度是个老大难问题——Cerema AWP数据集的发布正是为此而来：它在一条可以人工制造雨、雾、夜晚的隧道里，对同一场景、同一批行人反复拍摄，生成了条件完全受控、标注极为详尽的行人图像数据集。作者将该数据集整理成机器学习友好的HDF5格式，并针对天气分类、行人识别、方向识别、边界框回归、自动编码器重建以及条件生成对抗网络（Conditional GAN）等多项任务跑出了卷积神经网络基准结果：天气、行人、衣着等分类任务轻松超过99%准确率，方向识别约80%，回归和图像生成任务则揭示出网络先学背景、再学行人细节的有趣规律。此外，作者还按照"数据集数据表（Datasheets for Datasets）"规范对该数据集进行了完整描述，以增强透明度与可复现性。这项工作的价值在于：在任何花哨方法上线之前先晒出公平的基准线，并推动数据集标准化，从而帮助整个社区在算法比较和结果评估上更加客观、诚实。
原文 arXiv:1806.04016;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1806.04016v1