model = dict(
type='Tracktor', # The name of the multiple object tracker
detector=dict(
# Please refer to https://github.com/open-mmlab/mmdetection/blob/master/docs/tutorials/config.md#an-example-of-mask-r-cnn for detailed comments of detector.
type='FasterRCNN',
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),
rpn_head=dict(
type='RPNHead',
in_channels=256,
feat_channels=256,
anchor_generator=dict(
type='AnchorGenerator',
scales=[8],
ratios=[0.5, 1.0, 2.0],
strides=[4, 8, 16, 32, 64]),
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', use_sigmoid=True, loss_weight=1.0),
loss_bbox=dict(
type='SmoothL1Loss', beta=0.1111111111111111,
loss_weight=1.0)),
roi_head=dict(
type='StandardRoIHead',
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(
type='RoIAlign', output_size=7, sampling_ratio=0),
out_channels=256,
featmap_strides=[4, 8, 16, 32]),
bbox_head=dict(
type='Shared2FCBBoxHead',
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, 0.0, 0.0],
target_stds=[0.1, 0.1, 0.2, 0.2],
clip_border=False),
reg_class_agnostic=False,
loss_cls=dict(
type='CrossEntropyLoss',
use_sigmoid=False,
loss_weight=1.0),
loss_bbox=dict(type='SmoothL1Loss', loss_weight=1.0))),
init_cfg=dict(
type='Pretrained',
checkpoint= # noqa: E251
'https://download.openmmlab.com/mmtracking/mot/faster_rcnn/faster-rcnn_r50_fpn_4e_mot17-half-64ee2ed4.pth'
# noqa: E501
), # The pretrained weights of detector. It may also used for testing.
train_cfg=dict(
rpn=dict(
assigner=dict(
type='MaxIoUAssigner',
pos_iou_thr=0.7,
neg_iou_thr=0.3,
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_across_levels=False,
nms_pre=2000,
nms_post=1000,
max_num=1000,
nms_thr=0.7,
min_bbox_size=0),
rcnn=dict(
assigner=dict(
type='MaxIoUAssigner',
pos_iou_thr=0.5,
neg_iou_thr=0.5,
min_pos_iou=0.5,
match_low_quality=False,
ignore_iof_thr=-1),
sampler=dict(
type='RandomSampler',
num=512,
pos_fraction=0.25,
neg_pos_ub=-1,
add_gt_as_proposals=True),
pos_weight=-1,
debug=False)),
test_cfg=dict(
rpn=dict(
nms_across_levels=False,
nms_pre=1000,
nms_post=1000,
max_num=1000,
nms_thr=0.7,
min_bbox_size=0),
rcnn=dict(
score_thr=0.05,
nms=dict(type='nms', iou_threshold=0.5),
max_per_img=100))),
reid=dict( # The config of the ReID model
type='BaseReID', # The name of the motion model
backbone=dict( # The config of the backbone of the ReID model
type='ResNet',
# The type of the backbone, refer to https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/backbones/resnet.py#L288 for more details.
depth=50, # The depth of backbone, usually it is 50 or 101 for ResNet and ResNext backbones.
num_stages=4, # Number of stages of the backbone.
out_indices=(3,), # The index of output feature maps produced in each stages
style='pytorch'),
# The style of backbone, 'pytorch' means that stride 2 layers are in 3x3 conv, 'caffe' means stride 2 layers are in 1x1 convs.
neck=dict(type='GlobalAveragePooling', kernel_size=(8, 4), stride=1),
# The config of the neck of the ReID model. Generally it is a global average pooling module.
head=dict( # The config of the head of the ReID model.
type='LinearReIDHead', # The type of the classification head
num_fcs=1, # Number of the fully-connected layers in the head
in_channels=2048, # The number of the input channels
fc_channels=1024, # The number of channels of fc layers
out_channels=128, # The number of the output channels
norm_cfg=dict(type='BN1d'), # The config of the normalization modules
act_cfg=dict(type='ReLU')), # The config of the activation modules
init_cfg=dict(
type='Pretrained',
checkpoint= # noqa: E251
'https://download.openmmlab.com/mmtracking/mot/reid/reid_r50_6e_mot17-4bf6b63d.pth' # noqa: E501
)), # The pretrained weights of reid model. It may also used for testing.
motion=dict( # The config of the motion model
type='CameraMotionCompensation', # The name of the motion model
warp_mode='cv2.MOTION_EUCLIDEAN', # The warping mode
num_iters=100, # The number of the iterations
stop_eps=1e-05), # The threshold of termination
tracker=dict( # The config of the tracker
type='TracktorTracker', # The name of the tracker
obj_score_thr=0.5, # The score threshold to filter the detected objects
regression=dict( # The config of the regression part in Tracktor
obj_score_thr=0.5, # The score threshold to filter the regressed objects
nms=dict(type='nms', iou_threshold=0.6), # The nms config to filter the regressed objects
match_iou_thr=0.3), # The IoU threshold to filter the detected objects
reid=dict( # The config about the testing process of the ReID model
num_samples=10, # The maximum number of samples to calculate the feature embeddings
img_scale=(256, 128), # The input scale of the ReID model
img_norm_cfg=None,
# The normalization config of the input of the ReID model. None means consistent with the backbone
match_score_thr=2.0, # The threshold for feature similarity
match_iou_thr=0.2), # The threshold for IoU matching
momentums=None, # The momentums to update the buffers
num_frames_retain=10)) # The maximum number of frames to retain disappeared tracks
# The configs below are consistent with video object tracking. Please refer to `config_vid.md` for details.
dataset_type = 'MOTChallengeDataset'
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='LoadMultiImagesFromFile', to_float32=True),
dict(type='SeqLoadAnnotations', with_bbox=True, with_track=True),
dict(
type='SeqResize',
img_scale=(1088, 1088),
share_params=True,
ratio_range=(0.8, 1.2),
keep_ratio=True,
bbox_clip_border=False),
dict(type='SeqPhotoMetricDistortion', share_params=True),
dict(
type='SeqRandomCrop',
share_params=False,
crop_size=(1088, 1088),
bbox_clip_border=False),
dict(type='SeqRandomFlip', share_params=True, flip_ratio=0.5),
dict(
type='SeqNormalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='SeqPad', size_divisor=32),
dict(type='MatchInstances', skip_nomatch=True),
dict(
type='VideoCollect',
keys=[
'img', 'gt_bboxes', 'gt_labels', 'gt_match_indices',
'gt_instance_ids'
]),
dict(type='SeqDefaultFormatBundle', ref_prefix='ref')
]
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(1088, 1088),
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='VideoCollect', keys=['img'])
])
]
data_root = 'data/MOT17/'
data = dict(
samples_per_gpu=2,
workers_per_gpu=2,
train=dict(
type='MOTChallengeDataset',
visibility_thr=-1,
ann_file='data/MOT17/annotations/train_cocoformat.json',
img_prefix='data/MOT17/train',
ref_img_sampler=dict(
num_ref_imgs=1,
frame_range=10,
filter_key_img=True,
method='uniform'),
pipeline=train_pipeline),
val=dict(
type='MOTChallengeDataset',
ann_file='data/MOT17/annotations/train_cocoformat.json',
img_prefix='data/MOT17/train',
ref_img_sampler=None,
pipeline=test_pipeline),
test=dict(
type='MOTChallengeDataset',
ann_file='data/MOT17/annotations/train_cocoformat.json',
img_prefix='data/MOT17/train',
ref_img_sampler=None,
pipeline=test_pipeline))
optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=None)
checkpoint_config = dict(interval=1)
log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook', by_epoch=False),
dict(type='TensorboardLoggerHook', by_epoch=False),
dict(type='WandbLoggerHook', by_epoch=False,
init_kwargs={'entity': "OpenMMLab",
'project': "MMTracking",
'config': cfg_dict}),
])
dist_params = dict(backend='nccl', port='29500')
log_level = 'INFO'
load_from = None
resume_from = None
workflow = [('train', 1)]
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=100,
warmup_ratio=0.01,
step=[3])
total_epochs = 4
evaluation = dict(metric=['bbox', 'track'], interval=1)
search_metrics = ['MOTA', 'IDF1', 'FN', 'FP', 'IDs', 'MT', 'ML']
test_set = 'train'
work_dir = None