Market-1501-C 损坏图像数据集

Market-1501-C 是一个由应用于 Market-1501 测试集的算法生成的变体组成的评估数据集。由于在智能安全和视频监控等方面的广泛应用，人像识别已经成为计算机视觉领域的一个热门话题。然而将目前的人像识别模型部署到现实世界中仍有许多问题。与物体分类和检测不同，人像重新识别（ReID）是一个实例级的识别和排名问题，依赖于强大的和详细的信息。但目前的神经网络很容易被各种形式的破坏所损坏，如噪音、模糊等。该数据集包含的即是损坏图像，以帮助完成针对损坏图像学习不变的表示的任务。该数据集全面建立了 5 个 ReID 基准，用于学习再认定基准。

Market-1501-C is an evaluation dataset composed of variants generated by algorithms applied to the original Market-1501 test set. Person recognition has emerged as a prominent research topic in the field of computer vision due to its wide applications in intelligent security, video surveillance and other scenarios. However, deploying current person recognition models in real-world scenarios still faces numerous challenges. Unlike object classification and detection tasks, person re-identification (ReID) is an instance-level recognition and ranking problem that relies on robust and detailed feature information. Nevertheless, current neural networks are highly vulnerable to various types of corruptions, such as noise, blurring, and so forth. The dataset consists of these corrupted images to facilitate the task of learning invariant representations for images under various corruptions. This dataset comprehensively establishes 5 ReID benchmarks for person re-identification evaluation.

搜集汇总

背景与挑战

背景概述

Market-1501-C是一个由算法生成的损坏图像数据集，基于Market-1501测试集构建，包含噪声和模糊等损坏形式。它用于评估人像重新识别模型在现实世界损坏情况下的鲁棒性，并建立了5个ReID基准以学习对损坏图像不变的表示。