| dataset name | application scenarios for dataset | size of dataset and number of images | traffic related label types |
can or cannot dowmload directly | data type |
|---|---|---|---|---|---|
| COCO | pedestrian detection | 1.5 million application examples | phone/pedestrian | can | image |
| VOC | vhicle detection | 9963（VOC2007) | sedan/bus | can | image |
| BIT Vehicle | vhicle detection | 9850 | bus/truck/sedan /van/SUV |
Suspected download link is not available | image |
| KITTI | lane detection | 180GB | pedestrian/lane | can | image |
| Waymo | vhicle detection /Pedestrian detection |
200000 frames | vehicle/pedestrian /rider/sign |
can | video |
| Vehicle-1M | vhicle detection | 936051 | vehicle model | need to send a protocol | image |
| BerkeleyDeepDrive | lane detection /objects(people、cars、lights etc.)detection |
1100 hours of video/100，000 images | lane line | can | video image |
| VMMRdb | vehicle type recognition | 291752 | vehicle type | can | image |
| uulm | signal light detection | 230000+ | signal light | can（need to register） | image |
| UFPR | vehicle type and license plate detection | 4500 | seban/motor /license plate |
need to send a application email | image |
| CCPD | license plate detection for Chinese vehicles | 300,000+ | license plate's border and number |
can | image |
| Open Images | image classification /object detection/visual relationship detection |
9178275 | object classification /frame/visual relationship |
can（dowmload from CVDF） | image |
| BOSCH | lane line detection | 100,000+ | lane line | can（need to register） | image |
| Apollo Scape | lane line detection | 228GB | lane/zebra crossing /ground traffic sign |
can | image |
| PedX | Pedestrian detection | 255GB | frame/direction | can | image |
| VERI-Wild | vehicle detection | 30×24hours of video/40，671 images | vehicle type | can | video image |
| ECP Dataset | Pedestrian detection | 47300 | pedestrian/cycle /motor/SUV |
need to send application mail and related certificates | image |
| GOT-10k | object detection | 10,000 | motion state | need to send application mail | video |
| SUPERVISELY | AI auxiliary label | ||||
| CULane Dataset | lane line detection | 133,235 | lane line | can | image |
| Mapillary Vistas | object (all objects appearing in the picture) identification | 25,000 | People,vehicle, lane line, traffic light etc. (152 categories) | Need to request | image |
| HCI | traffic light detection (anomaly detection) | 13,427 | traffic light | can（need to fill in a form） | image |
| BOSCH | vehicle detection | 200000 | car | can（need to register） | image |
| Traffic Research | vehicle detection/license plate recognition/speed detection | 116k+2G | vehicle location/license plate recognition | can | image |
| Lyft | street dataset/vehicle identification | 55000 frames | lane/crosswalk/park zone/spatial semantic map | can（need to register） | image |
| Oxford Robotcar Dataset | street dataset | 10G~320G | road | can | image |
| nuScenes | street dataset/vehicle detection/Pedestrian detection | 1,400,000 | car/peopel/road | can（need to register） | image |
| HD1K Benchmark Suite | street dataset | 1000 frames | lane/traffic scene | need to send application mail | image |
| comma2k19 | street dataset/lane detection | 100G | more than 33 hours commute dataset on the highway | can | image |
| ILSVRC2017 | object detection |