Weakly Supervised Person Re-ID: Differentiable Graphical Learning and a New Benchmark

Person reidentification (Re-ID) benefits greatly from the accurate annotations of existing data sets (e.g., CUHK03 and Market-1501), which are quite expensive because each image in these has to be assigned with a proper label. In this work, we ease annotation Re-ID by replacing inaccurate annotation, i.e., group images into bags terms time assign bag-level label for bag. This reduces effort leads creation large-scale benchmark called SYSU- 30k. The new contains 30k individuals, is about 20 times larger than (1.3k individuals) Market-1501 (1.5k individuals), 30 ImageNet (1k categories). It sums up 29606918 images. Learning model weakly supervised problem. To solve problem, introduce differentiable graphical capture dependencies all bag generate reliable pseudolabel person's image. further used supervise learning model. Compared fully models, our method achieves state-of-the-art performance on other sets. code, set, pretrained will available at https://github.com/wanggrun/SYSU-30k.