Spaces:
Runtime error
Runtime error
| from .mask_rcnn_R_50_FPN_100ep_LSJ import ( | |
| dataloader, | |
| lr_multiplier, | |
| model, | |
| optimizer, | |
| train, | |
| ) | |
| from detectron2.config import LazyCall as L | |
| from detectron2.modeling.backbone import RegNet | |
| from detectron2.modeling.backbone.regnet import SimpleStem, ResBottleneckBlock | |
| # Config source: | |
| # https://github.com/facebookresearch/detectron2/blob/master/configs/COCO-InstanceSegmentation/mask_rcnn_regnetx_4gf_dds_fpn_1x.py # noqa | |
| model.backbone.bottom_up = L(RegNet)( | |
| stem_class=SimpleStem, | |
| stem_width=32, | |
| block_class=ResBottleneckBlock, | |
| depth=23, | |
| w_a=38.65, | |
| w_0=96, | |
| w_m=2.43, | |
| group_width=40, | |
| norm="SyncBN", | |
| out_features=["s1", "s2", "s3", "s4"], | |
| ) | |
| model.pixel_std = [57.375, 57.120, 58.395] | |
| # RegNets benefit from enabling cudnn benchmark mode | |
| train.cudnn_benchmark = True | |