-
Notifications
You must be signed in to change notification settings - Fork 2
/
mask2former_swin-s-p4-w7-224_lsj_4x1_50e_coco-person-finetune-SCP-5e.py
214 lines (198 loc) · 6.96 KB
/
mask2former_swin-s-p4-w7-224_lsj_4x1_50e_coco-person-finetune-SCP-5e.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
_base_ = ['./mask2former_swin-s-p4-w7-224_lsj_8x2_50e_coco.py']
classes = ('person',)
num_things_classes = len(classes)
# num_things_classes = 80
num_stuff_classes = 0
num_classes = num_things_classes + num_stuff_classes
load_from = "weights/mask2former_swin-s-p4-w7-224_lsj_8x2_50e_coco_20220504_001756-743b7d99.pth"
model = dict(
panoptic_head=dict(
num_things_classes=num_things_classes,
num_stuff_classes=num_stuff_classes,
loss_cls=dict(class_weight=[1.0] * num_classes + [0.1])),
panoptic_fusion_head=dict(
num_things_classes=num_things_classes,
num_stuff_classes=num_stuff_classes),
test_cfg=dict(panoptic_on=False))
# dataset settings
image_size = (1024, 1024)
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
pad_cfg = dict(img=(128, 128, 128), masks=0, seg=255)
load_pipeline = [
dict(type='LoadImageFromFile', to_float32=True),
dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
# large scale jittering
dict(
type='Resize',
img_scale=image_size,
ratio_range=(0.1, 2.0),
multiscale_mode='range',
keep_ratio=True),
dict(
type='RandomCrop',
crop_size=image_size,
crop_type='absolute',
recompute_bbox=True,
allow_negative_crop=True),
dict(type='FilterAnnotations', min_gt_bbox_wh=(1e-5, 1e-5), by_mask=True),
dict(type='RandomFlip', flip_ratio=0.5),
dict(type='Pad', size=image_size, pad_val=pad_cfg),
]
train_pipeline = [
dict(type='CopyPaste', max_num_pasted=100),
dict(type='Normalize', **img_norm_cfg),
dict(type='DefaultFormatBundle', img_to_float=True),
dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
]
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='Pad', size_divisor=32, pad_val=pad_cfg),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img']),
])
]
dataset_type = 'CocoDataset'
data_root = 'data/'
data = dict(
samples_per_gpu=1,
workers_per_gpu=1,
train=dict(
_delete_=True,
type='MultiImageMixDataset',
pipeline=train_pipeline,
dataset=dict(
type=dataset_type,
ann_file=data_root + 'COCO2017/annotations/instances_train2017.json',
img_prefix=data_root + 'COCO2017/train2017/',
classes=classes,
pipeline=load_pipeline,
),
),
val=dict(
type='ConcatDataset',
datasets=[
dict(
type=dataset_type,
ann_file=data_root + 'COCO2017/annotations/instances_val2017.json',
img_prefix=data_root + 'COCO2017/val2017/',
test_mode=True,
classes=classes,
pipeline=test_pipeline,
),
dict(
type=dataset_type,
ann_file=data_root + 'OCHuman/ochuman_coco_format_val_range_0.00_1.00_full_labelled.json',
img_prefix=data_root + 'OCHuman/images/',
classes=classes,
test_mode=True,
pipeline=test_pipeline,
),
]
),
test=dict(
type='ConcatDataset',
datasets=[
dict(
type=dataset_type,
ann_file=data_root + 'COCO2017/annotations/instances_val2017.json',
img_prefix=data_root + 'COCO2017/val2017/',
classes=classes,
test_mode=True,
pipeline=test_pipeline
),
dict(
type=dataset_type,
ann_file=data_root + 'OCHuman/ochuman_coco_format_val_range_0.00_1.00.json',
img_prefix=data_root + 'OCHuman/images/',
classes=classes,
test_mode=True,
pipeline=test_pipeline,
),
dict(
type=dataset_type,
ann_file=data_root + 'OCHuman/ochuman_coco_format_test_range_0.00_1.00.json',
img_prefix=data_root + 'OCHuman/images/',
classes=classes,
test_mode=True,
pipeline=test_pipeline,
),
dict(type=dataset_type,
ann_file=data_root + 'COCO2017/annotations/instances_val_person2017.json',
img_prefix=data_root + 'COCO2017/val2017/',
classes=classes,
test_mode=True,
pipeline=test_pipeline,
),
dict(
type=dataset_type,
ann_file=data_root + 'OCHuman/ochuman_coco_format_val_range_0.00_1.00_full_labelled.json',
img_prefix=data_root + 'OCHuman/images/',
classes=classes,
test_mode=True,
pipeline=test_pipeline,
),
dict(
type=dataset_type,
ann_file=data_root + 'OCHuman/ochuman_coco_format_test_range_0.00_1.00_full_labelled.json',
img_prefix=data_root + 'OCHuman/images/',
classes=classes,
test_mode=True,
pipeline=test_pipeline
),
]
)
)
# optimizer
optimizer = dict(
lr=0.0001/4/10, #adjusted for 20 epochs @ BS4, default was 1e-4 for BS16; Adjust further down by 10x due to unstable/high grad norms
)
# learning policy
lr_config = dict(
policy='step',
gamma=0.1,
by_epoch=False,
step=[71240, 77180], #adjusted for 5e@BS4
warmup='linear',
warmup_by_epoch=False,
warmup_ratio=1.0, # no warmup
warmup_iters=10)
max_iters = 80145 #adjusted for 5e@BS4
runner = dict(type='IterBasedRunner', max_iters=max_iters)
log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook', by_epoch=False),
dict(type='TensorboardLoggerHook', by_epoch=False),
# dict(
# type='WandbLoggerHook',
# init_kwargs=dict(
# project='ocp',
# name='m2f_swins_person-finetune-SCP-5e'
# ),
# out_suffix=('.log.json', '.log', '.py'),
# by_epoch=False,
# ),
]
)
interval = 16000
workflow = [('train', interval)]
checkpoint_config = dict(
by_epoch=False, interval=interval, save_last=True, max_keep_ckpts=3)
# Before 365001th iteration, we do evaluation every 5000 iterations.
# After 365000th iteration, we do evaluation every 368750 iterations,
# which means that we do evaluation at the end of training.
dynamic_intervals = [(max_iters // interval * interval + 1, max_iters)]
evaluation = dict(
interval=interval,
dynamic_intervals=dynamic_intervals,
save_best='1_segm_mAP',
metric=['bbox', 'segm'])