- we use
mmdetection
to train all models. - All models were trained on
bdd100k_train
, and tested on thebdd100k_val
. - We use distributed training across 8 Nvdia-1080Ti GPUs.
Name | backbone | tricks | AP | AP50 | AP75 | APs | APm | APl |
---|---|---|---|---|---|---|---|---|
FasterRCNN | R_50_FPN | 0.318 | 0.551 | 0.311 | 0.145 | 0.356 | 0.497 | |
FasterRCNN | R_101_FPN | 0.322 | 0.553 | 0.314 | 0.142 | 0.360 | 0.512 | |
CascadeRCNN | R_50_FPN | 0.332 | 0.558 | 0.331 | 0.150 | 0.371 | 0.520 | |
PISA | R_50_FPN | |||||||
LibraRCNN | R_50_FPN | |||||||
GA | R_50_FPN |
Name | backbone | tricks | AP | AP50 | AP75 | APs | APm | APl |
---|---|---|---|---|---|---|---|---|
FCOS | R_50_FPN | 0.304 | 0.539 | 0.290 | 0.129 | 0.338 | 0.498 | |
ATSS | R_50_FPN | 0.329 | 0.562 | 0.323 | 0.141 | 0.367 | 0.517 | |
CenterNet | R_50_DCN | |||||||
RepPoints | R_50_FPN | 0.312 | 0.555 | 0.297 | 0.129 | 0.348 | 0.505 |
Name | backbone | Iters | AP | AP50 | AP75 | APs | APm | APl |
---|---|---|---|---|---|---|---|---|
CenterNet | R_50_DCN | 125997 | 27.5269 | 44.7613 | 28.8301 | 9.6805 | 31.4682 | 43.1641 |