No segmentation result after single-class sample training, boundary box support only

[Mia]

Hello, I build a new single-class dataset, training based on r101-refinemask-2x, and the modified configs are as follows.

However, after 93 epochs, I was stuck in some trouble. On the one hand, I found the detection results had no segmentation mask, and on the other hand, loss_instance was really difficult to converge. For example, {"mode": "train", "epoch": 94, "iter": 10, "lr": 3e-05, "time": 0.57583, "data_time": 0.2215, "memory": 5419, "loss_rpn_cls": 0.00174, "loss_rpn_bbox": 0.00334, "loss_cls": 0.01584, "acc": 99.6875, "loss_bbox": 0.03414, "loss_instance": 0.26567, "loss_semantic": 0.00546, "loss": 0.3262, "grad_norm": 2.08605}

I have two questions now:
1. Does refineMask support instance segmentation of elongated objects, or is there something wrong with my configuration file?
2. Any recommended epochs of training?
Thanks for your time.

model = dict(
type='MaskRCNN',
pretrained='torchvision://resnet101',
backbone=dict(
type='ResNet',
depth=101,
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='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]),
loss_cls=dict(
type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
roi_head=dict(
type='RefineRoIHead',
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]),
reg_class_agnostic=False,
loss_cls=dict(
type='CrossEntropyLoss', use_sigmoid=False, loss_weight=2.0),
loss_bbox=dict(type='L1Loss', loss_weight=2.0)),
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', output_size=14, sampling_ratio=0),
out_channels=256,
featmap_strides=[4, 8, 16, 32]),
mask_head=dict(
type='RefineMaskHead',
num_convs_instance=2,
num_convs_semantic=4,
conv_in_channels_instance=256,
conv_in_channels_semantic=256,
conv_kernel_size_instance=3,
conv_kernel_size_semantic=3,
conv_out_channels_instance=256,
conv_out_channels_semantic=256,
conv_cfg=None,
norm_cfg=None,
dilations=[1, 3, 5],
semantic_out_stride=4,
mask_use_sigmoid=True,
stage_num_classes=[1, 1, 1, 1],
stage_sup_size=[14, 28, 56, 112],
upsample_cfg=dict(type='bilinear', scale_factor=2),
loss_cfg=dict(
type='RefineCrossEntropyLoss',
stage_instance_loss_weight=[0.25, 0.5, 0.75, 1.0],
semantic_loss_weight=1.0,
boundary_width=2,
start_stage=1))))
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=True,
ignore_iof_thr=-1),
sampler=dict(
type='RandomSampler',
num=512,
pos_fraction=0.25,
neg_pos_ub=-1,
add_gt_as_proposals=True),
mask_size=28,
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,
mask_thr_binary=0.5))
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', to_float32=True),
dict(
type='LoadAnnotations',
with_bbox=True,
with_mask=True,
poly2mask=False),
dict(type='Resize', img_scale=(1024, 1024), keep_ratio=True),
dict(type='RandomFlip', flip_ratio=0.5),
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='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks'])
]
test_pipeline = [
dict(type='LoadImageFromFile', to_float32=True),
dict(
type='MultiScaleFlipAug',
img_scale=(1024, 1024),
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='Collect', keys=['img'])
])
]
data_root = '../coco'
data = dict(
samples_per_gpu=1,
workers_per_gpu=2,
train=dict(
type='CocoDataset',
ann_file='annotations/instances_train2017.json',
img_prefix='train2017',
pipeline=[
dict(type='LoadImageFromFile', to_float32=True),
dict(
type='LoadAnnotations',
with_bbox=True,
with_mask=True,
poly2mask=False),
dict(type='Resize', img_scale=(1024, 1024), keep_ratio=True),
dict(type='RandomFlip', flip_ratio=0.5),
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='DefaultFormatBundle'),
dict(
type='Collect',
keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks'])
],
data_root='../coco',
classes=('street', )),
val=dict(
type='CocoDataset',
ann_file='annotations/instances_train2017.json',
img_prefix='train2017',
pipeline=[
dict(type='LoadImageFromFile', to_float32=True),
dict(
type='MultiScaleFlipAug',
img_scale=(1024, 1024),
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='Collect', keys=['img'])
])
],
data_root='../coco',
classes=('street', )),
test=dict(
type='CocoDataset',
ann_file='annotations/instances_train2017.json',
img_prefix='train2017',
pipeline=[
dict(type='LoadImageFromFile', to_float32=True),
dict(
type='MultiScaleFlipAug',
img_scale=(1024, 1024),
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='Collect', keys=['img'])
])
],
data_root='../coco',
classes=('street', )))
evaluation = dict(metric=['bbox', 'segm'], classwise=True, interval=12)
optimizer = dict(type='SGD', lr=0.003, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
total_epochs = 1000
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=0.001,
step=[16, 22])
checkpoint_config = dict(interval=1)
log_config = dict(interval=10, hooks=[dict(type='TextLoggerHook')])
dist_params = dict(backend='nccl')
log_level = 'INFO'
workflow = [('train', 1)]
gpu_ids = range(0, 1)
work_dir = 'work_dirs/r101-refinemask-2x/street'
load_from = None
resume_from = 'work_dirs/r101-refinemask-2x/street/latest.pth'
classes = ('street', )

[Gang Zhang]

I am very sorry to forget to reply to you. Hope you have solved this problem.