crosskd_r50_gflv1_r101-2x-ms_fpn_1x_coco.py

_base_ = [
    '../_base_/datasets/coco_detection.py',
    '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
]
teacher_ckpt = 'https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r101_fpn_mstrain_2x_coco/gfl_r101_fpn_mstrain_2x_coco_20200629_200126-dd12f847.pth'  # noqa
model = dict(
    type='CrossKDGFL',
    data_preprocessor=dict(
        type='DetDataPreprocessor',
        mean=[123.675, 116.28, 103.53],
        std=[58.395, 57.12, 57.375],
        bgr_to_rgb=True,
        pad_size_divisor=32),
    teacher_config='configs/gfl/gfl_r101_fpn_ms-2x_coco.py',
    teacher_ckpt=teacher_ckpt,
    backbone=dict(
        type='ResNet',
        depth=50,
        num_stages=4,
        out_indices=(0, 1, 2, 3),
        frozen_stages=1,
        norm_cfg=dict(type='BN', requires_grad=True),
        norm_eval=True,
        style='pytorch',
        init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
    neck=dict(
        type='FPN',
        in_channels=[256, 512, 1024, 2048],
        out_channels=256,
        start_level=1,
        add_extra_convs='on_output',
        num_outs=5),
    bbox_head=dict(
        type='GFLHead',
        num_classes=80,
        in_channels=256,
        stacked_convs=4,
        feat_channels=256,
        anchor_generator=dict(
            type='AnchorGenerator',
            ratios=[1.0],
            octave_base_scale=8,
            scales_per_octave=1,
            strides=[8, 16, 32, 64, 128]),
        loss_cls=dict(
            type='QualityFocalLoss',
            use_sigmoid=True,
            beta=2.0,
            loss_weight=1.0),
        loss_dfl=dict(type='DistributionFocalLoss', loss_weight=0.25),
        reg_max=16,
        loss_bbox=dict(type='GIoULoss', loss_weight=2.0)),
    kd_cfg=dict(
        loss_cls_kd=dict(type='KDQualityFocalLoss', beta=1, loss_weight=1.0),
        loss_reg_kd=dict(
            type='KnowledgeDistillationKLDivLoss',
            class_reduction='sum',
            T=1,
            loss_weight=4.0),
        reused_teacher_head_idx=3),
    # model training and testing settings
    train_cfg=dict(
        assigner=dict(type='ATSSAssigner', topk=9),
        allowed_border=-1,
        pos_weight=-1,
        debug=False),
    test_cfg=dict(
        nms_pre=1000,
        min_bbox_size=0,
        score_thr=0.05,
        nms=dict(type='nms', iou_threshold=0.6),
        max_per_img=100))

# optimizer
optim_wrapper = dict(
    type='OptimWrapper',
    optimizer=dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001))

train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=12, val_interval=1)
default_hooks = dict(checkpoint=dict(type='CheckpointHook', interval=12))
train_dataloader = dict(batch_size=2, num_workers=4)
auto_scale_lr = dict(enable=True, base_batch_size=16)