centernet-update_r50-caffe_fpn_ms-1x_coco.py

_base_ = [
    '../_base_/datasets/coco_detection.py',
    '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
]

model = dict(
    type='CenterNet',
    # use caffe img_norm
    data_preprocessor=dict(
        type='DetDataPreprocessor',
        mean=[103.530, 116.280, 123.675],
        std=[1.0, 1.0, 1.0],
        bgr_to_rgb=False,
        pad_size_divisor=32),
    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=False),
        norm_eval=True,
        style='caffe',
        init_cfg=dict(
            type='Pretrained',
            checkpoint='open-mmlab://detectron2/resnet50_caffe')),
    neck=dict(
        type='FPN',
        in_channels=[256, 512, 1024, 2048],
        out_channels=256,
        start_level=1,
        add_extra_convs='on_output',
        num_outs=5,
        # There is a chance to get 40.3 after switching init_cfg,
        # otherwise it is about 39.9~40.1
        init_cfg=dict(type='Caffe2Xavier', layer='Conv2d'),
        relu_before_extra_convs=True),
    bbox_head=dict(
        type='CenterNetUpdateHead',
        num_classes=80,
        in_channels=256,
        stacked_convs=4,
        feat_channels=256,
        strides=[8, 16, 32, 64, 128],
        hm_min_radius=4,
        hm_min_overlap=0.8,
        more_pos_thresh=0.2,
        more_pos_topk=9,
        soft_weight_on_reg=False,
        loss_cls=dict(
            type='GaussianFocalLoss',
            pos_weight=0.25,
            neg_weight=0.75,
            loss_weight=1.0),
        loss_bbox=dict(type='GIoULoss', loss_weight=2.0),
    ),
    train_cfg=None,
    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))

# single-scale training is about 39.3
train_pipeline = [
    dict(type='LoadImageFromFile', backend_args={{_base_.backend_args}}),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(
        type='RandomChoiceResize',
        scales=[(1333, 640), (1333, 672), (1333, 704), (1333, 736),
                (1333, 768), (1333, 800)],
        keep_ratio=True),
    dict(type='RandomFlip', prob=0.5),
    dict(type='PackDetInputs')
]

train_dataloader = dict(dataset=dict(pipeline=train_pipeline))

# learning rate
param_scheduler = [
    dict(
        type='LinearLR',
        start_factor=0.00025,
        by_epoch=False,
        begin=0,
        end=4000),
    dict(
        type='MultiStepLR',
        begin=0,
        end=12,
        by_epoch=True,
        milestones=[8, 11],
        gamma=0.1)
]

optim_wrapper = dict(
    optimizer=dict(lr=0.01),
    # Experiments show that there is no need to turn on clip_grad.
    paramwise_cfg=dict(norm_decay_mult=0.))

# NOTE: `auto_scale_lr` is for automatically scaling LR,
# USER SHOULD NOT CHANGE ITS VALUES.
# base_batch_size = (8 GPUs) x (2 samples per GPU)
auto_scale_lr = dict(base_batch_size=16)