mmdetection/configs/crowddet/crowddet-rcnn_r50_fpn_8xb2-30e_crowdhuman.py

_base_ = ['../_base_/default_runtime.py']

model = dict(
    type='CrowdDet',
    data_preprocessor=dict(
        type='DetDataPreprocessor',
        mean=[103.53, 116.28, 123.675],
        std=[57.375, 57.12, 58.395],
        bgr_to_rgb=False,
        pad_size_divisor=64,
        # This option is set according to https://github.com/Purkialo/CrowdDet/
        # blob/master/lib/data/CrowdHuman.py The images in the entire batch are
        # resize together.
        batch_augments=[
            dict(type='BatchResize', scale=(1400, 800), pad_size_divisor=64)
        ]),
    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,
        num_outs=5,
        upsample_cfg=dict(mode='bilinear', align_corners=False)),
    rpn_head=dict(
        type='RPNHead',
        in_channels=256,
        feat_channels=256,
        anchor_generator=dict(
            type='AnchorGenerator',
            scales=[8],
            ratios=[1.0, 2.0, 3.0],
            strides=[4, 8, 16, 32, 64],
            centers=[(8, 8), (8, 8), (8, 8), (8, 8), (8, 8)]),
        bbox_coder=dict(
            type='DeltaXYWHBBoxCoder',
            target_means=[0.0, 0.0, 0.0, 0.0],
            target_stds=[1.0, 1.0, 1.0, 1.0],
            clip_border=False),
        loss_cls=dict(type='CrossEntropyLoss', loss_weight=1.0),
        loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
    roi_head=dict(
        type='MultiInstanceRoIHead',
        bbox_roi_extractor=dict(
            type='SingleRoIExtractor',
            roi_layer=dict(
                type='RoIAlign',
                output_size=7,
                sampling_ratio=-1,
                aligned=True,
                use_torchvision=True),
            out_channels=256,
            featmap_strides=[4, 8, 16, 32]),
        bbox_head=dict(
            type='MultiInstanceBBoxHead',
            with_refine=False,
            num_shared_fcs=2,
            in_channels=256,
            fc_out_channels=1024,
            roi_feat_size=7,
            num_classes=1,
            bbox_coder=dict(
                type='DeltaXYWHBBoxCoder',
                target_means=[0., 0., 0., 0.],
                target_stds=[0.1, 0.1, 0.2, 0.2]),
            reg_class_agnostic=False,
            loss_cls=dict(
                type='CrossEntropyLoss',
                loss_weight=1.0,
                use_sigmoid=False,
                reduction='none'),
            loss_bbox=dict(
                type='SmoothL1Loss', loss_weight=1.0, reduction='none'))),
    # model training and testing settings
    train_cfg=dict(
        rpn=dict(
            assigner=dict(
                type='MaxIoUAssigner',
                pos_iou_thr=0.7,
                neg_iou_thr=(0.3, 0.7),
                min_pos_iou=0.3,
                match_low_quality=True,
                ignore_iof_thr=-1),
            sampler=dict(
                type='RandomSampler',
                num=256,
                pos_fraction=0.5,
                neg_pos_ub=-1,
                add_gt_as_proposals=False),
            allowed_border=-1,
            pos_weight=-1,
            debug=False),
        rpn_proposal=dict(
            nms_pre=2400,
            max_per_img=2000,
            nms=dict(type='nms', iou_threshold=0.7),
            min_bbox_size=2),
        rcnn=dict(
            assigner=dict(
                type='MultiInstanceAssigner',
                pos_iou_thr=0.5,
                neg_iou_thr=0.5,
                min_pos_iou=0.3,
                match_low_quality=False,
                ignore_iof_thr=-1),
            sampler=dict(
                type='MultiInsRandomSampler',
                num=512,
                pos_fraction=0.5,
                neg_pos_ub=-1,
                add_gt_as_proposals=False),
            pos_weight=-1,
            debug=False)),
    test_cfg=dict(
        rpn=dict(
            nms_pre=1200,
            max_per_img=1000,
            nms=dict(type='nms', iou_threshold=0.7),
            min_bbox_size=2),
        rcnn=dict(
            nms=dict(type='nms', iou_threshold=0.5),
            score_thr=0.01,
            max_per_img=500)))

dataset_type = 'CrowdHumanDataset'
data_root = 'data/CrowdHuman/'

# Example to use different file client
# Method 1: simply set the data root and let the file I/O module
# automatically infer from prefix (not support LMDB and Memcache yet)

# data_root = 's3://openmmlab/datasets/tracking/CrowdHuman/'

# Method 2: Use `backend_args`, `file_client_args` in versions before 3.0.0rc6
# backend_args = dict(
#     backend='petrel',
#     path_mapping=dict({
#         './data/': 's3://openmmlab/datasets/tracking/',
#         'data/': 's3://openmmlab/datasets/tracking/'
#     }))
backend_args = None

train_pipeline = [
    dict(type='LoadImageFromFile', backend_args=backend_args),
    dict(type='LoadAnnotations', with_bbox=True),
    dict(type='RandomFlip', prob=0.5),
    dict(
        type='PackDetInputs',
        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
                   'flip_direction'))
]
test_pipeline = [
    dict(type='LoadImageFromFile', backend_args=backend_args),
    dict(type='Resize', scale=(1400, 800), keep_ratio=True),
    # avoid bboxes being resized
    dict(type='LoadAnnotations', with_bbox=True),
    dict(
        type='PackDetInputs',
        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
                   'scale_factor'))
]

train_dataloader = dict(
    batch_size=2,
    num_workers=4,
    persistent_workers=True,
    sampler=dict(type='DefaultSampler', shuffle=True),
    batch_sampler=None,  # The 'batch_sampler' may decrease the precision
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file='annotation_train.odgt',
        data_prefix=dict(img='Images/'),
        filter_cfg=dict(filter_empty_gt=True, min_size=32),
        pipeline=train_pipeline,
        backend_args=backend_args))
val_dataloader = dict(
    batch_size=1,
    num_workers=2,
    persistent_workers=True,
    drop_last=False,
    sampler=dict(type='DefaultSampler', shuffle=False),
    dataset=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file='annotation_val.odgt',
        data_prefix=dict(img='Images/'),
        test_mode=True,
        pipeline=test_pipeline,
        backend_args=backend_args))
test_dataloader = val_dataloader

val_evaluator = dict(
    type='CrowdHumanMetric',
    ann_file=data_root + 'annotation_val.odgt',
    metric=['AP', 'MR', 'JI'],
    backend_args=backend_args)
test_evaluator = val_evaluator

train_cfg = dict(type='EpochBasedTrainLoop', max_epochs=30, val_interval=1)
val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop')
param_scheduler = [
    dict(
        type='LinearLR', start_factor=0.001, by_epoch=False, begin=0, end=800),
    dict(
        type='MultiStepLR',
        begin=0,
        end=30,
        by_epoch=True,
        milestones=[24, 27],
        gamma=0.1)
]

# optimizer
auto_scale_lr = dict(base_batch_size=16)
optim_wrapper = dict(
    type='OptimWrapper',
    optimizer=dict(type='SGD', lr=0.002, momentum=0.9, weight_decay=0.0001))