Exert Diversity and Mitigate Bias: Domain Generalizable Person Re-identification with a Comprehensive Benchmark
Datasets | Collected Scenes | # Identities | # Cameras | # Total Images |
---|---|---|---|---|
Small-Scale | ||||
iLIDS | Airport Arrival Hall | 119 | 2 | 476 |
GRID | Underground Station | 1,025 | 8 | 1,275 |
PKU | Campus | 114 | 2 | 1,824 |
SAIVT | Buildings | 152 | 8 | 7,150 |
Medium-Scale | ||||
Market-1501 | Campus | 1,501 | 6 | 29,419 |
LPW | Street | 2,731 | 4 | 30,678 |
WildTrack | An Open Area | 313 | 7 | 33,979 |
Airport | Airport | 9,651 | 6 | 39,902 |
Large-Scale | ||||
SOMAset | Synthetic | 50 | - | 100,000 |
Unreal | Synthetic | 1,960 | 34 | 119,128 |
MSMT17 | Indoor and Outdoor | 4,101 | 15 | 126,441 |
PersonX | Synthetic | 1,266 | 6 | 273,456 |
- CUDA>=10.0
- Four 1080-Ti GPUs
- necessary packages listed in requirements.txt
This repo borrows partially from fast-reid