CODA Benchmark
*COMMON evaluates 5 frequent road object categories in CODA including pedestrian, cyclist, car, truck, and tricycle. *NOVEL evaluates the rest 29 categories in (CODA Base + CODA-ONCE) in a class-agnostic manner. *CORNER evaluates all 34 categories in (CODA Base + CODA-ONCE), including both COMMON and NOVEL, in a class-agnostic manner.
| Method | Dataset | CORNER | COMMON | NOVEL | ||||||||||||
| mAR | AR50 | AR75 | AR1 | AR10 | mAR | AR50 | AR75 | AR1 | AR10 | mAR | AR50 | AR75 | AR1 | AR10 | ||
| CODA-ONCE | ||||||||||||||||
| RetinaNet | SODA10M | 9.6 | 21.8 | 7.0 | 0.2 | 3.0 | 29.8 | 59.8 | 24.5 | 4.7 | 24.8 | - | - | - | - | - |
| Faster R-CNN | 4.0 | 8.5 | 3.4 | 0.2 | 2.3 | 23.3 | 44.3 | 19.6 | 3.7 | 22.5 | - | - | - | - | - | |
| Cascade R-CNN | 5.6 | 11.4 | 4.9 | 0.2 | 2.9 | 27.2 | 46.0 | 30.7 | 4.9 | 25.3 | - | - | - | - | - | |
| RPN (Faster) | 5.1 | 11.3 | 4.0 | 0.3 | 1.3 | - | - | - | - | - | - | - | - | - | - | |
| RPN (Cascade) | 4.8 | 11.4 | 3.5 | 0.3 | 1.3 | - | - | - | - | - | - | - | - | - | - | |
| ORE | 5.3 | 11.5 | 4.4 | 0.3 | 2.8 | 19.2 | 36.8 | 17.1 | 5.4 | 18.6 | 2.0 | 4.8 | 1.6 | 0.4 | 1.7 | |
| RetinaNet | BDD100K | 10.2 | 19.4 | 9.1 | 0.2 | 2.6 | 30.1 | 63.7 | 23.3 | 5.9 | 25.8 | 7.3 | 14.1 | 6.6 | 0.6 | 3.8 |
| Faster R-CNN | 7.8 | 15.1 | 7.4 | 0.1 | 2.3 | 26.8 | 51.8 | 23.2 | 4.7 | 24.3 | 5.5 | 10.7 | 5.2 | 0.5 | 4.1 | |
| Cascade R-CNN | 7.7 | 14.6 | 6.9 | 0.1 | 2.4 | 26.7 | 48.8 | 24.0 | 5.4 | 24.6 | 5.7 | 10.5 | 5.1 | 0.6 | 4.3 | |
| RPN (Faster) | 7.9 | 16.1 | 7.0 | 0.2 | 1.6 | - | - | - | - | - | - | - | - | - | - | |
| RPN (Cascade) | 7.7 | 15.8 | 6.9 | 0.2 | 1.8 | - | - | - | - | - | - | - | - | - | - | |
| Faster R-CNN | Waymo | 4.5 | 9.5 | 3.6 | 0.2 | 2.7 | 17.5 | 32.9 | 15.1 | 2.8 | 17.0 | 2.5 | 5.3 | 2.0 | 0.1 | 1.9 |
| RPN (Faster) | 5.0 | 10.8 | 4.3 | 0.2 | 1.7 | - | - | - | - | - | - | - | - | - | - | |
| CODA Base + CODA-ONCE | ||||||||||||||||
| RetinaNet | SODA10M | 11.9 | 25.2 | 9.5 | 0.2 | 5.4 | 28.7 | 58.9 | 23.5 | 5.0 | 23.9 | - | - | - | - | - |
| Faster R-CNN | 6.8 | 13.0 | 6.4 | 0.2 | 4.9 | 23.9 | 46.8 | 20.1 | 5.6 | 23.1 | - | - | - | - | - | |
| Cascade R-CNN | 8.3 | 15.5 | 7.6 | 0.2 | 5.5 | 27.2 | 47.0 | 29.4 | 6.2 | 25.3 | - | - | - | - | - | |
| RPN (Faster) | 8.1 | 16.2 | 7.4 | 0.6 | 3.1 | - | - | - | - | - | - | - | - | - | - | |
| RPN (Cascade) | 7.7 | 16.0 | 6.8 | 0.5 | 2.8 | - | - | - | - | - | - | - | - | - | - | |
| ORE | 8.3 | 16.4 | 7.4 | 0.3 | 5.6 | 18.5 | 35.5 | 18.2 | 5.5 | 18.1 | 3.4 | 7.6 | 2.8 | 0.8 | 2.9 | |
| RetinaNet | BDD100K | 12.8 | 23.2 | 11.9 | 0.1 | 4.8 | 27.5 | 58.1 | 21.5 | 5.4 | 23.6 | 9.7 | 17.7 | 9.1 | 1.2 | 5.9 |
| Faster R-CNN | 10.7 | 19.2 | 10.2 | 0.1 | 4.3 | 24.4 | 48.1 | 20.9 | 4.3 | 22.0 | 7.2 | 13.3 | 6.8 | 0.9 | 5.9 | |
| Cascade R-CNN | 10.4 | 18.5 | 9.7 | 0.1 | 4.5 | 25.7 | 48.4 | 23.3 | 5.8 | 23.6 | 6.9 | 12.5 | 6.5 | 1.1 | 5.7 | |
| RPN (Faster) | 10.6 | 20.0 | 10.2 | 0.5 | 3.7 | - | - | - | - | - | - | - | - | - | - | |
| RPN (Cascade) | 10.6 | 20.0 | 10.2 | 0.4 | 3.9 | - | - | - | - | - | - | - | - | - | - | |
| Faster R-CNN | Waymo | 6.1 | 11.9 | 5.4 | 0.2 | 4.4 | 16.3 | 30.5 | 13.9 | 2.8 | 15.7 | 4.4 | 8.6 | 4.1 | 0.1 | 3.7 |
| RPN (Faster) | 6.7 | 13.4 | 6.4 | 0.2 | 2.6 | - | - | - | - | - | - | - | - | - | - | |