adamixer_r50_1x_coco.py

def __get_debug():
    import os
    return 'C_DEBUG' in os.environ


debug = __get_debug()

log_interval = 100


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

IMAGE_SCALE = (1333, 800)

# dataset settings
dataset_type = 'CocoDataset'
data_root = 'data/coco/'
img_norm_cfg = dict(
    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(type='Resize', img_scale=IMAGE_SCALE, keep_ratio=True),
    dict(type='RandomFlip', flip_ratio=0.5),
    dict(type='Normalize', **img_norm_cfg),
    dict(type='Pad', size_divisor=32),
    dict(type='DefaultFormatBundle'),
    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),
]
test_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(
        type='MultiScaleFlipAug',
        img_scale=IMAGE_SCALE,
        flip=False,
        transforms=[
            dict(type='Resize', keep_ratio=True),
            dict(type='RandomFlip'),
            dict(type='Normalize', **img_norm_cfg),
            dict(type='Pad', size_divisor=32),
            dict(type='ImageToTensor', keys=['img']),
            dict(type='Collect', keys=['img']),
        ])
]
data = dict(
    samples_per_gpu=2,
    workers_per_gpu=2,
    train=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_train2017.json',
        img_prefix=data_root + 'train2017/',
        pipeline=train_pipeline),
    val=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/instances_val2017.json',
        img_prefix=data_root + 'val2017/',
        pipeline=test_pipeline),
    test=dict(
        type=dataset_type,
        ann_file=data_root + 'annotations/image_info_test-dev2017.json',
        img_prefix=data_root + 'test2017/',
        pipeline=test_pipeline),
)
evaluation = dict(interval=1, metric='bbox')


num_stages = 6
num_query = 100
QUERY_DIM = 256
FEAT_DIM = 256
FF_DIM = 2048

# P_in for spatial mixing in the paper.
in_points_list = [32, ] * num_stages

# P_out for spatial mixing in the paper. Also named as `out_points` in this codebase.
out_patterns_list = [128, ] * num_stages

# G for the mixer grouping in the paper. Please distinguishe it from num_heads in MHSA in this codebase.
n_group_list = [4, ] * num_stages

model = dict(
    type='QueryBased',
    pretrained='torchvision://resnet50',
    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'),
    neck=dict(
        type='ChannelMapping',
        in_channels=[256, 512, 1024, 2048],
        out_channels=FEAT_DIM,
        start_level=0,
        add_extra_convs='on_output',
        num_outs=4),
    rpn_head=dict(
        type='InitialQueryGenerator',
        num_query=num_query,
        content_dim=QUERY_DIM),
    roi_head=dict(
        type='AdaMixerDecoder',
        featmap_strides=[4, 8, 16, 32],
        num_stages=num_stages,
        stage_loss_weights=[1] * num_stages,
        content_dim=QUERY_DIM,
        bbox_head=[
            dict(
                type='AdaMixerDecoderStage',
                num_classes=80,
                num_ffn_fcs=2,
                num_heads=8,
                num_cls_fcs=1,
                num_reg_fcs=1,
                feedforward_channels=FF_DIM,
                content_dim=QUERY_DIM,
                feat_channels=FEAT_DIM,
                dropout=0.0,
                in_points=in_points_list[stage_idx],
                out_points=out_patterns_list[stage_idx],
                n_groups=n_group_list[stage_idx],
                ffn_act_cfg=dict(type='ReLU', inplace=True),
                loss_bbox=dict(type='L1Loss', loss_weight=5.0),
                loss_iou=dict(type='GIoULoss', loss_weight=2.0),
                loss_cls=dict(
                    type='FocalLoss',
                    use_sigmoid=True,
                    gamma=2.0,
                    alpha=0.25,
                    loss_weight=2.0),
                # NOTE: The following argument is a placeholder to hack the code. No real effects for decoding or updating bounding boxes.
                bbox_coder=dict(
                    type='DeltaXYWHBBoxCoder',
                    clip_border=False,
                    target_means=[0., 0., 0., 0.],
                    target_stds=[0.5, 0.5, 1., 1.])) for stage_idx in range(num_stages)
        ]),
    # training and testing settings
    train_cfg=dict(
        rpn=None,
        rcnn=[
            dict(
                assigner=dict(
                    type='HungarianAssigner',
                    cls_cost=dict(type='FocalLossCost', weight=2.0),
                    reg_cost=dict(type='BBoxL1Cost', weight=5.0),
                    iou_cost=dict(type='IoUCost', iou_mode='giou',
                                  weight=2.0)),
                sampler=dict(type='PseudoSampler'),
                pos_weight=1) for _ in range(num_stages)
        ]),
    test_cfg=dict(rpn=None, rcnn=dict(max_per_img=num_query)))

# optimizer
optimizer = dict(
    _delete_=True,
    type='AdamW',
    lr=0.000025,
    weight_decay=0.0001,
)

optimizer_config = dict(
    _delete_=True, grad_clip=dict(max_norm=1.0, norm_type=2),
)

# learning policy
lr_config = dict(
    policy='step',
    step=[8, 11],
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=0.001
)
runner = dict(type='EpochBasedRunner', max_epochs=12)


def __date():
    import datetime
    return datetime.datetime.now().strftime('%m%d_%H%M')


log_config = dict(
    interval=log_interval,
    hooks=[
        dict(type='TextLoggerHook'),
        # dict(type='TensorboardLoggerHook')
    ]
)

postfix = '_' + __date()

find_unused_parameters = True


resume_from = None