# We follow the original implementation which
# adopts the Caffe pre-trained backbone.
_base_ = [
    '../_base_/datasets/coco_detection.py',
    '../_base_/schedules/schedule_1x.py', '../_base_/default_runtime.py'
]
# model settings
model = dict(
    type='AutoAssign',
    data_preprocessor=dict(
        type='DetDataPreprocessor',
        mean=[102.9801, 115.9465, 122.7717],
        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=True,
        num_outs=5,
        relu_before_extra_convs=True,
        init_cfg=dict(type='Caffe2Xavier', layer='Conv2d')),
    bbox_head=dict(
        type='AutoAssignHead',
        num_classes=80,
        in_channels=256,
        stacked_convs=4,
        feat_channels=256,
        strides=[8, 16, 32, 64, 128],
        loss_bbox=dict(type='GIoULoss', loss_weight=5.0)),
    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))

# learning rate
param_scheduler = [
    dict(
        type='LinearLR', start_factor=0.001, by_epoch=False, begin=0,
        end=1000),
    dict(
        type='MultiStepLR',
        begin=0,
        end=12,
        by_epoch=True,
        milestones=[8, 11],
        gamma=0.1)
]

# optimizer
optim_wrapper = dict(
    optimizer=dict(lr=0.01), paramwise_cfg=dict(norm_decay_mult=0.))