跨摄像头检索
1. 提取行人特征（除脸部特征以外），存入特征数据库。
2. 对新的图像进行重识别，进行特征匹配
深度学习自提取特征。数据集有：
1) VIPeR
2) PRID2011
3) CUHK03
4) Market1501
1. "bounding_box_test" file. This file contains 19732 images for testing. 用于测试
2. "bounding_box_train" file. This file contains 12936 images for training. 用于训练
3. "query" file. It contains 3368 query images. Search is performed in the "bounding_box_test" file.查询文件。包括3368张查询图片。用于在1测试test集中查询。
4. "gt_bbox" file. It contains 25259 images, all hand-drawn. The images correspond to all the 1501 individuals in the test and training set.It is used to distinguish "good" "junk" and "distractors".它包含2万多张图片，手画的。对应的是1501个人。用来区分”好“，”差“和“干扰”。
5. "gt_query" file. For each of the 3368 queries, there are both good and junk relevant images (containing the same individual). This file contains the image index of the good and junk images. It is used during performance evaluation.对于3368个查询，有好和差的相关图片（对于同样的一个人）。该文件包括好的和差的图片的索引。它是用于执行评价阶段的。