# model settings
model = dict(
    type='SOLOv2',
    backbone=dict(
        type='MobileNetV2',
        #num_stages=4,
        #out_indices=(0, 1, 2, 3), # C2, C3, C4, C5
        out_indices=(1,2,4,7),
        frozen_stages=1,
        #style='pytorch'
        init_cfg=dict(type='Pretrained', 
                      #checkpoint='torchvision://mobilenet_v2'
                      checkpoint='mobilenet_v2-b0353104.pth'
                     ),
    ),
    neck=dict(
        type='FPN',
        in_channels=[24,32,64,320], #[24,32,96,1280]
        out_channels=8,
        start_level=0,
        num_outs=5),
    mask_head=dict(
        type='SOLOV2Head',
        num_classes=1,#80
        in_channels=8,
        feat_channels=32,# can not modify
        stacked_convs=4,
        strides=[8, 8, 16, 32, 32],
        scale_ranges=((1, 56), (28, 112), (56, 224), (112, 448), (224, 896)),
        pos_scale=0.2,
        num_grids=[40, 36, 24, 16, 12],
        cls_down_index=0,
        mask_feature_head=dict(
            feat_channels=8,
            start_level=0,
            end_level=3,
            out_channels=8,
            mask_stride=4,
            norm_cfg=dict(type='GN', num_groups=8, requires_grad=True)),#32,16
        loss_mask=dict(type='DiceLoss', use_sigmoid=True, loss_weight=3.0),
        loss_cls=dict(
            type='FocalLoss',
            use_sigmoid=True,
            gamma=2.0,
            alpha=0.25,
            loss_weight=1.0)),
    test_cfg=dict(
        nms_pre=500,
        score_thr=0.1,
        mask_thr=0.5,
        filter_thr=0.05,
        kernel='gaussian',
        sigma=2.0,
        max_per_img=100))

data = dict(
    samples_per_gpu=6,
    workers_per_gpu=6,
    train=dict(
        type='CocoDataset',
        ann_file='/dataset/coco/annotations/instances_train2017.json',
        img_prefix='/dataset/coco/train2017/images',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
            dict(
                type='Resize',
                img_scale=[(768, 512), (768, 480), (768, 448), (768, 416), 
                           (768, 384), (768, 352)],
                multiscale_mode='value',
                keep_ratio=True),
            dict(type='RandomFlip', flip_ratio=0.5),
            dict(
                type='Normalize',
                mean=[123.675, 116.28, 103.53],
                std=[58.395, 57.12, 57.375],
                to_rgb=True),
            dict(type='Pad', size_divisor=32),
            dict(type='DefaultFormatBundle'),
            dict(
                type='Collect',
                keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks'])
        ]),
    val=dict(
        type='CocoDataset',
        ann_file='/dataset/coco/annotations/instances_val2017.json',
        img_prefix='/dataset/coco/val2017/images',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(768, 448),
                flip=False,
                transforms=[
                    dict(type='Resize', keep_ratio=True),
                    dict(type='RandomFlip'),
                    dict(
                        type='Normalize',
                        mean=[123.675, 116.28, 103.53],
                        std=[58.395, 57.12, 57.375],
                        to_rgb=True),
                    dict(type='Pad', size_divisor=32),
                    dict(type='ImageToTensor', keys=['img']),
                    dict(type='Collect', keys=['img'])
                ])
        ]),
    test=dict(
        type='CocoDataset',
        ann_file = '/data/person_data/model_test_portrait_1.json',
        img_prefix='/',
        pipeline=[
            dict(type='LoadImageFromFile'),
            dict(
                type='MultiScaleFlipAug',
                img_scale=(768, 448),
                flip=False,
                transforms=[
                    dict(type='Resize', keep_ratio=True),
                    dict(type='RandomFlip'),
                    dict(
                        type='Normalize',
                        mean=[123.675, 116.28, 103.53],
                        std=[58.395, 57.12, 57.375],
                        to_rgb=True),
                    dict(type='Pad', size_divisor=32),
                    dict(type='ImageToTensor', keys=['img']),
                    dict(type='Collect', keys=['img'])
                ])
        ]))
evaluation = dict(metric=['bbox', 'segm'])
optimizer = dict(type='SGD', lr=0.01, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=0.3333333333333333,
    step=[20, 25, 31] #[27, 33]
)
runner = dict(type='EpochBasedRunner', max_epochs=36)
checkpoint_config = dict(interval=1)
log_config = dict(interval=50, hooks=[dict(type='TextLoggerHook')])
custom_hooks = [dict(type='NumClassCheckHook')]
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = None
resume_from = None
workflow = [('train', 1)]
opencv_num_threads = 0
mp_start_method = 'fork'
auto_scale_lr = dict(enable=False, base_batch_size=16)

find_unused_parameters = True
temp = 0.5
alpha_fgd = 0.001
beta_fgd = 0.0005
gamma_fgd = 0.0005
lambda_fgd = 5e-06
distiller = dict(
    type='DetectionDistiller',
    teacher_pretrained='work_dirs/solov2_r101_3x_c8/epoch_36.pth',
    init_student=True,
    distill_cfg=[
        dict(
            student_module='neck.fpn_convs.3.conv',
            teacher_module='neck.fpn_convs.3.conv',
            output_hook=True,
            methods=[
                dict(
                    type='FeatureLoss',
                    name='loss_fgd_fpn_3',
                    student_channels=8,# 256
                    teacher_channels=8,# 256
                    temp=0.5,
                    alpha_fgd=0.001,
                    beta_fgd=0.0005,
                    gamma_fgd=0.0005,
                    lambda_fgd=5e-06)
            ]),
        dict(
            student_module='neck.fpn_convs.2.conv',
            teacher_module='neck.fpn_convs.2.conv',
            output_hook=True,
            methods=[
                dict(
                    type='FeatureLoss',
                    name='loss_fgd_fpn_2',
                    student_channels=8,
                    teacher_channels=8,
                    temp=0.5,
                    alpha_fgd=0.001,
                    beta_fgd=0.0005,
                    gamma_fgd=0.0005,
                    lambda_fgd=5e-06)
            ]),
        dict(
            student_module='neck.fpn_convs.1.conv',
            teacher_module='neck.fpn_convs.1.conv',
            output_hook=True,
            methods=[
                dict(
                    type='FeatureLoss',
                    name='loss_fgd_fpn_1',
                    student_channels=8,
                    teacher_channels=8,
                    temp=0.5,
                    alpha_fgd=0.001,
                    beta_fgd=0.0005,
                    gamma_fgd=0.0005,
                    lambda_fgd=5e-06)
            ]),
        dict(
            student_module='neck.fpn_convs.0.conv',
            teacher_module='neck.fpn_convs.0.conv',
            output_hook=True,
            methods=[
                dict(
                    type='FeatureLoss',
                    name='loss_fgd_fpn_0',
                    student_channels=8,
                    teacher_channels=8,
                    temp=0.5,
                    alpha_fgd=0.001,
                    beta_fgd=0.0005,
                    gamma_fgd=0.0005,
                    lambda_fgd=5e-06)
            ])
    ])
student_cfg = 'configs/solov2/solov2_mobileNetV2_3x_c8.py'
teacher_cfg = 'configs/solov2/solov2_r101_3x_c8.py'
work_dir = './work_dirs/fgd_solov2_r101_c8_3x_ms_distill_solov2_mobileNetV2_c8_3x'
auto_resume = False
gpu_ids = [0]