Knowledge Distillataion for mmdetection model 2

[kdh-awraw1019]

base = [
'../../base/datasets/mmdet/coco_detection.py',
'../../base/schedules/mmdet/schedule_1x.py',
'../../base/mmdet_runtime.py'
]

##################### model settings
student = dict(
type='mmdet.YOLOF',
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='DilatedEncoder',
in_channels=2048,
out_channels=512,
block_mid_channels=128,
num_residual_blocks=4),
bbox_head=dict(
type='YOLOFHead',
num_classes=2,
in_channels=512,
reg_decoded_bbox=True,
anchor_generator=dict(
type='AnchorGenerator',
ratios=[1.0],
scales=[1, 2, 4, 8, 16],
strides=[32]),
bbox_coder=dict(
type='DeltaXYWHBBoxCoder',
target_means=[.0, .0, .0, .0],
target_stds=[1., 1., 1., 1.],
add_ctr_clamp=True,
ctr_clamp=32),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=1.0),
loss_bbox=dict(type='GIoULoss', loss_weight=1.0)),
################## training and testing settings
train_cfg=dict(
assigner=dict(
type='UniformAssigner', pos_ignore_thr=0.15, neg_ignore_thr=0.7),
allowed_border=-1,
pos_weight=-1,
debug=False),
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))

teacher = dict(
type='mmdet.YOLOF',
backbone=dict(
type='ResNeXt',
depth=101,
groups=64,
base_width=4,
num_stages=4,
out_indices=(1, 2, 3),
frozen_stages=1,
norm_cfg=dict(type='BN', requires_grad=True),
style='pytorch',
init_cfg=None),
neck=dict(
type='DilatedEncoder',
in_channels=2048,
out_channels=512,
block_mid_channels=128,
num_residual_blocks=4),
bbox_head=dict(
type='YOLOFHead',
num_classes=2,
in_channels=512,
reg_decoded_bbox=True,
anchor_generator=dict(
type='AnchorGenerator',
ratios=[1.0],
scales=[1, 2, 4, 8, 16],
strides=[32]),
bbox_coder=dict(
type='DeltaXYWHBBoxCoder',
target_means=[.0, .0, .0, .0],
target_stds=[1., 1., 1., 1.],
add_ctr_clamp=True,
ctr_clamp=32),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=1.0),
loss_bbox=dict(type='GIoULoss', loss_weight=1.0)),
################### training and testing settings
train_cfg=dict(
assigner=dict(
type='UniformAssigner', pos_ignore_thr=0.15, neg_ignore_thr=0.7),
allowed_border=-1,
pos_weight=-1,
debug=False),
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))

################### use caffe img_norm
img_norm_cfg = dict(
mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True),
dict(type='Resize', img_scale=(1333, 800), keep_ratio=True),
dict(type='RandomFlip', flip_ratio=0.5),
dict(type='RandomShift', shift_ratio=0.5, max_shift_px=32),
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=(1333, 800),
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']),
])
]

################ optimizer
################ optimizer by mmdetection documents
optimizer = dict(type='SGD', lr=1e-3, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict()

################### learning policy by mmdetection documents
lr_config = dict(policy='step',
min_lr=1e-3,
# by_epoch=True,
warmup='linear',
warmup_iters=5000,
warmup_ratio=1.0 / 5,
step=1,
warmup_by_epoch=False)
runner = dict(type='EpochBasedRunner', max_epochs=150)
evaluation= dict(classwise=True, interval=1, metric='bbox')

#################### algorithm setting
algorithm = dict(
type='GeneralDistill',
architecture=dict(
type='MMDetArchitecture',
model=student,
),
distiller=dict(
type='SingleTeacherDistiller',
teacher=teacher,
teacher_trainable=False,
components=[
dict(
student_module='bbox_head.cls_score',
teacher_module='bbox_head.cls_score',
losses=[
dict(
type='ChannelWiseDivergence',
name='loss_cwd_cls_head',
tau=1,
loss_weight=5,
)
])
]),
)

find_unused_parameters = True
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)]

################

python tools/mmdet/train_mmdet.py ./configs/distill/cwd/cbox_cwd_cls_head_yolof_r101_fpn_yolof_r50_fpn_1x_coco.py --cfg-options algorithm.distiller.teacher.init_cfg.type=Pretrained algorithm.distiller.teacher.init_cfg.checkpoint=./teacher_model/yolof/epoch_115.pth

################# error message

Traceback (most recent call last):
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 52, in build_from_cfg
return obj_cls(**args)
TypeError: init() got an unexpected keyword argument 'init_cfg'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 52, in build_from_cfg
return obj_cls(**args)
File "/home/.../mmrazor/mmrazor/models/distillers/single_teacher.py", line 35, in init
self.teacher = self.build_teacher(teacher)
File "/home/.../mmrazor/mmrazor/models/distillers/single_teacher.py", line 71, in build_teacher
teacher = MODELS.build(cfg)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 212, in build
return self.build_func(*args, **kwargs, registry=self)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/cnn/builder.py", line 27, in build_model_from_cfg
return build_from_cfg(cfg, registry, default_args)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 55, in build_from_cfg
raise type(e)(f'{obj_cls.name}: {e}')
TypeError: YOLOF: init() got an unexpected keyword argument 'init_cfg'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 52, in build_from_cfg
return obj_cls(**args)
File "/home/.../mmrazor/mmrazor/models/algorithms/general_distill.py", line 23, in init
super(GeneralDistill, self).init(**kwargs)
File "/home/.../mmrazor/mmrazor/models/algorithms/base.py", line 57, in init
self._init_distiller(distiller)
File "/home/.../mmrazor/mmrazor/models/algorithms/base.py", line 134, in _init_distiller
self.distiller = build_distiller(distiller)
File "/home/.../mmrazor/mmrazor/models/builder.py", line 35, in build_distiller
return DISTILLERS.build(cfg)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 212, in build
return self.build_func(*args, **kwargs, registry=self)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/cnn/builder.py", line 27, in build_model_from_cfg
return build_from_cfg(cfg, registry, default_args)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 55, in build_from_cfg
raise type(e)(f'{obj_cls.name}: {e}')
TypeError: SingleTeacherDistiller: YOLOF: init() got an unexpected keyword argument 'init_cfg'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "tools/mmdet/train_mmdet.py", line 204, in
main()
File "tools/mmdet/train_mmdet.py", line 177, in main
algorithm = build_algorithm(cfg.algorithm)
File "/home/.../mmrazor/mmrazor/models/builder.py", line 20, in build_algorithm
return ALGORITHMS.build(cfg)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 212, in build
return self.build_func(*args, **kwargs, registry=self)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/cnn/builder.py", line 27, in build_model_from_cfg
return build_from_cfg(cfg, registry, default_args)
File "/opt/conda/envs/mmrazor/lib/python3.7/site-packages/mmcv/utils/registry.py", line 55, in build_from_cfg
raise type(e)(f'{obj_cls.name}: {e}')
TypeError: GeneralDistill: SingleTeacherDistiller: YOLOF: init() got an unexpected keyword argument 'init_cfg'

#################################

[pppppM]

What is your mmdet version？

[kdh-awraw1019]

My setting is

PyTorch: 1.7.0
TorchVision: 0.8.0
OpenCV: 4.5.5
MMCV: 1.4.2
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.0
MMDetection: 2.22.0+81e0e34
mmrazor: 0.2.0

[pppppM]

Because YOLOF is not updated in time, init_cfg cannot be used.
I will contact the developers of mmdet to fix this as soon as possible.

The quick solution is to add init_cfg to the args of YOLOF, and you can refer to YOLOX.
https://github.com/open-mmlab/mmdetection/blob/bde7b4b7eea9dd6ee91a486c6996b2d68662366d/mmdet/models/detectors/yolox.py#L18-L20

[kdh-awraw1019]

Thank you :)