mmdetection/configs/masktrack_rcnn/masktrack-rcnn_mask-rcnn_x101_fpn_8xb1-12e_youtubevis2021.py

_base_ = ['./masktrack-rcnn_mask-rcnn_r50_fpn_8xb1-12e_youtubevis2019.py']
model = dict(
    detector=dict(
        backbone=dict(
            type='ResNeXt',
            depth=101,
            groups=64,
            base_width=4,
            init_cfg=dict(
                type='Pretrained',
                checkpoint='open-mmlab://resnext101_64x4d')),
        init_cfg=dict(
            type='Pretrained',
            checkpoint=  # noqa: E251
            'https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_x101_64x4d_fpn_1x_coco/mask_rcnn_x101_64x4d_fpn_1x_coco_20200201-9352eb0d.pth'  # noqa: E501
        )))

data_root = 'data/youtube_vis_2021/'
dataset_version = data_root[-5:-1]

# dataloader
train_dataloader = dict(
    dataset=dict(
        data_root=data_root,
        dataset_version=dataset_version,
        ann_file='annotations/youtube_vis_2021_train.json'))
val_dataloader = dict(
    dataset=dict(
        data_root=data_root,
        dataset_version=dataset_version,
        ann_file='annotations/youtube_vis_2021_valid.json'))
test_dataloader = val_dataloader