SYSU-30k数据集，全球最大的ReID数据集

Posted 2023-01-23 AI浩

文章目录

• 摘要
• 与现有的Re-ID数据集进行对比
• 数据集下载链接
• 数据集结构
• 测试结果

摘要

论文链接：https://arxiv.org/abs/1904.03845
中山大学发布了一个大型行人重识别数据集也诞生了，即 SYSU-30k。SYSU-30k 数据集包含 30,000 个行人身份类别，约是 CUHK03 和 Market-1501 的 20 倍。如果一个行人身份类别相当于一个物体类别的话，则 SYSU-30k 相当于 ImageNet 的 30 倍。该数据集总共包含 29,606,918 张图像。

与现有的Re-ID数据集进行对比

Dataset	CUHK03	Market-1501	Duke	MSMT17	CUHK01	PRID	VIPeR	CAVIAR	SYSU-30k
Categories	1,467	1,501.	1,812	4,101	971	934	632	72	30,508
Scene	Indoor	Outdoor	Outdoor	Indoor, Outdoor	Indoor	Outdoor	Outdoor	Indoor	Indoor, Outdoor
Annotation	Strong	Strong	Strong	Strong	Strong	Strong	Strong	Strong	Weak
Cameras	2	6	8	15	10	2	2	2	Countless
Images	28,192	32,668	36,411	126,441	3,884	1,134	1,264	610	29,606,918

数据集下载链接

数据集分别存在谷歌网盘和百度网盘，链接如下：

• 谷歌网盘：https://drive.google.com/drive/folders/1MTxZ4UN_mbxjByZgcAki-H10zDzzeyuJ?usp=sharing
• 百度网盘：https://pan.baidu.com/s/1Y9phSZ5jy02szFZB_KqlyQ 提取码：1qzv

网盘中数据集的目录结构:

sysu-30k-released
├── train
|   ├── sysu_train_set_all_part1.tar
|   ├── sysu_train_set_all_part2.tar
|   ├── sysu_train_set_all_part3.tar
|   ├── sysu_train_set_all_part4.tar
|   ├── sysu_train_set_all_part5.tar
|   ├── sysu_train_set_all_part6.tar
|   ├── sysu_train_set_all_part7.tar
|   ├── sysu_train_set_all_part8.tar
|   ├── sysu_train_set_all_part9.tar
|   ├── sysu_train_set_all_part10.tar

├── test
|   ├── sysu_test_set_all.tar
├── train.txt (for weakly supervised training, "filename\\n" in each line)
├── query.txt (for evaluation)
├── gallery.txt (for evaluation)
├── net_6.pth (pretrained model)

数据集结构

解压后数据集的结构：

SYSU-30k-released
├── SYSU-30k-released
│   ├── meta
│   |   ├── train.txt (for weakly supervised training, "filename\\n" in each line)
│   |   ├── query.txt (for evaluation)
│   |   ├── gallery.txt (for evaluation)
│   ├── sysu_train_set_all
│   |   ├── 0000000001
│   |   ├── 0000000002
│   |   ├── 0000000003
│   |   ├── 0000000004
│   |   ├── ...
│   |   ├── 0000028309
│   |   ├── 0000028310
│   ├── sysu_test_set_all
│   |   ├── gallery
│   |   |   ├── 000028311
│   |   |   |   ├── 000028311_c1_1.jpg
│   |   |   ├── 000028312
│   |   |   |   ├── 000028312_c1_1.jpg
│   |   |   ├── 000028313Leaderboard on SYSU-30k.
│   |   |   |   ├── 000028313_c1_1.jpg
│   |   |   ├── 000028314
│   |   |   |   ├── 000028314_c1_1.jpg
│   |   |   ├── ...
│   |   |   |   ├── ...
│   |   |   ├── 000029309
│   |   |   |   ├── 000029309_c1_1.jpg
│   |   |   ├── 000029310
│   |   |   |   ├── 000029310_c1_1.jpg
│   |   |   ├── 0000others
│   |   |   |   ├── 0000others_c1_1.jpg
│   |   |   |   ├── ...
│   |   |   |   ├── ...
│   |   ├── query
│   |   |   ├── 000028311
│   |   |   |   ├── 000028311_c2_2.jpg
│   |   |   ├── 000028312
│   |   |   |   ├── 000028312_c2_2.jpg
│   |   |   ├── 000028313
│   |   |   |   ├── 000028313_c2_2.jpg
│   |   |   ├── 000028314
│   |   |   |   ├── 000028314_c2_2.jpg
│   |   |   ├── ...
│   |   |   |   ├── ...
│   |   |   ├── 000029309
│   |   |   |   ├── 000029309_c2_2.jpg
│   |   |   ├── 000029310
│   |   |   |   ├── 000029310_c2_2.jpg

测试结果

paperswithcode链接：https://paperswithcode.com/sota/person-re-identification-on-sysu-30k

Supervision	Method	Rank-1
Generalization	DARI [1]	11.2
Generalization	DF [2]	10.3
Generalization	ResNet-50 [3]	20.1
Generalization	Local-CNN [4]	23.0
Generalization	MGN [5]	23.6
Generalization	IICS [6]	36.0
Weakly Supervised	DGL [7]	26.9
Self Supervised	SimCLR [8]	10.9
Self Supervised	MoCo v2 [9]	11.6
Self Supervised	BYOL [10]	12.7
Self Supervised	Triplet [11]	14.8