新地标识别数据集
收藏资源简介:
新地标识别数据集由谷歌研究院创建,旨在提供一个公平的全球地标代表性。该数据集通过结合匿名的谷歌地图用户贡献统计和贡献者的 demographic 信息来估计地标的公平相关性,从而创建了一个比现有数据集更全面和公平的数据集。数据集包含50,000个地标,通过分层抽样方法,考虑了性别、年龄和国家居住地等 demographic 类别,以增加数据集的多样性。该数据集主要用于评估计算机视觉模型,特别是在2021年的谷歌地标识别和检索挑战中,以解决现有数据集可能存在的偏见问题。
The New Landmark Recognition Dataset was created by Google Research to achieve fair and representative global landmark coverage. To develop this dataset, researchers combined anonymized Google Maps user contribution statistics with contributors' demographic information to estimate the fair relevance of each landmark, resulting in a more comprehensive and equitable dataset than existing datasets. The dataset contains 50,000 landmarks, which were selected via stratified sampling that accounts for demographic categories including gender, age, and country of residence, to enhance the dataset's diversity. This dataset is primarily used for evaluating computer vision models, most notably in the 2021 Google Landmark Recognition and Retrieval Challenge, to address potential bias issues present in existing datasets.

- 1Towards A Fairer Landmark Recognition Dataset谷歌研究院 · 2022年



