Open douling843 opened 2 weeks ago
taugeren
commented
2 weeks ago Thank you, and I have already fixed it. You can set "visualize=True" in "train_config" to visualize the CPM during training, and you need to modify "store_dir" to a specific directory.
douling843
commented
2 weeks ago Thank you for your reply, after your fix this issue has been resolved, but a new problem has arisen. Since I have only one target category: [ship], I made the following changes: in train_cpm_ssdd.py ==========> base = [ '../base/datasets/ssdd.py', '../base/schedules/schedule_1x.py', '../base/default_runtime.py' ]
data_root = 'data/ssdd/'
classes = ('ship', )
in cpm_head.py ============> def get_mask_image(self, max_probs, max_indices, thr, num_width):
PALETTE = [
(0, 255, 0),
]
mask_image = np.ones((num_width, num_width, 3), dtype=np.uint8) * 255
for i in range(num_width):
for j in range(num_width):
if max_probs[i, j] > thr:
mask_image[i, j] = PALETTE[max_indices[i, j]]
return mask_image============================================ So the following error occurred:
2024-10-11 08:22:54,412 - mmrotate - INFO - workflow: [('train', 1)], max: 6 epochs
2024-10-11 08:22:54,412 - mmrotate - INFO - Checkpoints will be saved to /workspace/PointOBB-v2/work_dirs/cpm_ssdd by HardDiskBackend.
/opt/conda/envs/POBBv2/lib/python3.8/site-packages/torch/functional.py:445: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at ../aten/src/ATen/native/TensorShape.cpp:2157.)
return _VF.meshgrid(tensors, kwargs) # type: ignore[attr-defined]
Traceback (most recent call last):
File "tools/train.py", line 196, in
Please, how can I fix it?
taugeren
commented
2 weeks ago May I have a look on your config?
douling843
commented
2 weeks ago base = [ '../base/datasets/ssdd.py', '../base/schedules/schedule_1x.py', '../base/default_runtime.py' ]
data_root = 'data/ssdd/' ################################
store_dir = 'PointOBB-v2/your_absolute_vis_dir_ssdd/visualize' ##########################################PointOBB-v2/your_absolute_vis_dir_ssdd
angle_version = 'le90'
classes = ('ship', ) #####################################################
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'), dict(type='LoadAnnotations', with_bbox=True), dict(type='RResize', img_scale=(608, 608)), ################ dict(type='RResize', img_scale=(1024, 1024)), dict( type='RRandomFlip', flip_ratio=[0.25, 0.25, 0.25], direction=['horizontal', 'vertical', 'diagonal'], version=angle_version), dict(type='Normalize', **img_norm_cfg), dict(type='Pad', size_divisor=32), dict(type='DefaultFormatBundle'), dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']) ]
# data = dict(
# train=dict(pipeline=train_pipeline, version=angle_version),
# val=dict(version=angle_version),
# test=dict(version=angle_version))data = dict( train=dict( pipeline=train_pipeline, ann_file=data_root + 'train/labelTxt/', img_prefix=data_root + 'train/images/', version=angle_version, classes=classes), val=dict( ann_file=data_root + 'test/all/labelTxt/', img_prefix=data_root + 'test/all/images/', version=angle_version, classes=classes), test=dict( ann_file=data_root + 'test/all/labelTxt/', img_prefix=data_root + 'test/all/images/', version=angle_version, classes=classes, samples_per_gpu=4))
model = dict( type='RotatedFCOS', backbone=dict( type='ResNet', depth=50, num_stages=4, out_indices=(0, 1, 2, 3), frozen_stages=1, zero_init_residual=False, 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, start_level=0, add_extra_convs='on_output', # use P5 num_outs=6, relu_before_extra_convs=True),
bbox_head=dict(
type='CPMHead',
num_classes=len(classes),
in_channels=256,
stacked_convs=4,
feat_channels=256,
regress_ranges=((-1, 32), (32, 64), (64, 128), (128, 256), (256, 512),
(512, 1e8)),
strides=[4, 8, 16, 32, 64, 128],
center_sampling=True,
center_sample_radius=1.5,
norm_on_bbox=True,
centerness_on_reg=True,
separate_angle=False,
scale_angle=True,
bbox_coder=dict(
type='DistanceAnglePointCoder', angle_version=angle_version),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=1.0),
loss_bbox=dict(type='RotatedIoULoss', loss_weight=1.0),
loss_centerness=dict(
type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0)),
# training and testing settings
train_cfg=dict(
visualize=True, #######################################################visualize=False,
store_dir=store_dir,
cls_weight=1.0,
thresh1=6,
alpha=1.5
),
test_cfg=dict(
store_dir=store_dir,
nms_pre=2000,
min_bbox_size=0,
score_thr=0.05,
nms=dict(iou_thr=0.1),
max_per_img=2000))find_unused_parameters = True runner = dict(delete=True, type='EpochBasedRunner', max_epochs=6) lr_config = dict( delete=True, policy='step', warmup='linear', warmup_iters=500, warmup_ratio=1.0 / 3, step=[4]) evaluation = dict(interval=6, metric='mAP') optimizer = dict(lr=0.025) ##########################################
taugeren
commented
2 weeks ago Maybe you can try to set "dict(type='RResize', img_scale=(1024, 1024))" in "train_pipeline"
douling843
commented
2 weeks ago Maybe you can try to set "dict(type='RResize', img_scale=(1024, 1024))" in "train_pipeline"
I gradually tried to set the size of the image on the small side, and when setting img_scale=(64, 64), the code is working, but it's not appropriate for us to set the size of the image so small, and we get AP=0.000. since the number of categories in my dataset is 1(ship), I think it may be different from the DOTA dataset's number of categories, and a certain parameter needs to be changed, but I don't know exactly which parameter.
taugeren
commented
2 weeks ago Did img_scale=(1024, 1024) works? and its normal to have a AP=0.000 on train_cpm_stage. You need also to generate pseudo labels and train detectors.
douling843
commented
2 weeks ago Did img_scale=(1024, 1024) works? and its normal to have a AP=0.000 on train_cpm_stage. You need also to generate pseudo labels and train detectors.
when img_scale=(1024, 1024)，=======>>>> IndexError: index 184 is out of bounds for dimension 0 with size 184.
img_scale=(1024, 1024) doesn't work, but as (1024, 1024)--->(512, 512)--->(256, 256)--->(128, 128)--->(64, 64), until (64, 64) code will work.
Isn't Training CPM the first step? This is the step where this error occurs.
taugeren
commented
2 weeks ago Did img_scale=(1024, 1024) works? and its normal to have a AP=0.000 on train_cpm_stage. You need also to generate pseudo labels and train detectors.
when img_scale=(1024, 1024)，=======>>>> IndexError: index 184 is out of bounds for dimension 0 with size 184.
img_scale=(1024, 1024) doesn't work, but as (1024, 1024)--->(512, 512)--->(256, 256)--->(128, 128)--->(64, 64), until (64, 64) code will work.
but (64, 64) is too small for an image. The error is in visualization code, and what is the "max_probs" in this setting. It should be [h/self.stride[0], w/self.stride[0]] (where self.stride[0] is 4 in setting).
Isn't Training CPM the first step? This is the step where this error occurs.
It is normal that AP=0 during the phase of CPM training because the network don't know the boundary of the object. Only when using pseudo label to train detector, the AP is not zero.
douling843
commented
2 weeks ago Did img_scale=(1024, 1024) works? and its normal to have a AP=0.000 on train_cpm_stage. You need also to generate pseudo labels and train detectors.
when img_scale=(1024, 1024)，=======>>>> IndexError: index 184 is out of bounds for dimension 0 with size 184. img_scale=(1024, 1024) doesn't work, but as (1024, 1024)--->(512, 512)--->(256, 256)--->(128, 128)--->(64, 64), until (64, 64) code will work.
but (64, 64) is too small for an image. The error is in visualization code, and what is the "max_probs" in this setting. It should be [h/self.stride[0], w/self.stride[0]] (where self.stride[0] is 4 in setting).
Isn't Training CPM the first step? This is the step where this error occurs.
It is normal that AP=0 during the phase of CPM training because the network don't know the boundary of the object. Only when using pseudo label to train detector, the AP is not zero.
Thanks for your reply, but where should I find the visualisation code, ‘max_probs’ appears in cpm_head.py and I don't find [h/self.stride[0], w/self.stride[0]].
taugeren
commented
2 weeks ago Line 100~146 in "mmrotate/models/dense_heads/cpm_head.py"
douling843
commented
1 week ago Line 100~146 in "mmrotate/models/dense_heads/cpm_head.py"
max_probs in cpm_head.py is a variable or matrix, we are not forced to make its dimensions, and I don't understand where you say [h/self.stride[0], w/self.stride[0]]?
taugeren
commented
1 week ago File "/workspace/PointOBB-v2/mmrotate/models/dense_heads/cpm_head.py", line 116, in get_mask_image if max_probs[i, j] > thr: IndexError: index 120 is out of bounds for dimension 0 with size 120
The error report position, the max_probs, should be [h/self.stride[0], w/self.stride[0]], and it is [256, 256] in the DOTA setting.
I followed the installation tutorial to run the project, but there is an error, what should I do about this error?
2024-10-10 14:02:38,142 - mmrotate - INFO - Set random seed to 1029784185, deterministic: False Traceback (most recent call last): File "/opt/conda/envs/POBBv2/lib/python3.8/site-packages/mmcv/utils/registry.py", line 69, in build_from_cfg return obj_cls(**args) File "/workspace/PointOBB-v2/mmrotate/models/dense_heads/cpm_head.py", line 89, in init if kwargs.get('train_cfg')['vis_train_duration'] is not None: File "/opt/conda/envs/POBBv2/lib/python3.8/site-packages/mmcv/utils/config.py", line 40, in missing raise KeyError(name) KeyError: 'vis_train_duration'
During handling of the above exception, another exception occurred:
Traceback (most recent call last): File "/opt/conda/envs/POBBv2/lib/python3.8/site-packages/mmcv/utils/registry.py", line 69, in build_from_cfg return obj_cls(*args) File "/workspace/PointOBB-v2/mmrotate/models/detectors/rotated_fcos.py", line 21, in init super(RotatedFCOS, self).init(backbone, neck, bbox_head, train_cfg, File "/workspace/PointOBB-v2/mmrotate/models/detectors/single_stage.py", line 35, in init self.bbox_head = build_head(bbox_head) File "/workspace/PointOBB-v2/mmrotate/models/builder.py", line 37, in build_head return ROTATED_HEADS.build(cfg) File "/opt/conda/envs/POBBv2/lib/python3.8/site-packages/mmcv/utils/registry.py", line 237, in build return self.build_func(args, **kwargs, registry=self) File "/opt/conda/envs/POBBv2/lib/python3.8/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/POBBv2/lib/python3.8/site-packages/mmcv/utils/registry.py", line 72, in build_from_cfg raise type(e)(f'{obj_cls.name}: {e}') KeyError: "CPMHead: 'vis_train_duration'"
During handling of the above exception, another exception occurred:
Traceback (most recent call last): File "tools/train.py", line 196, in
main()
File "tools/train.py", line 165, in main
model = build_detector(
File "/workspace/PointOBB-v2/mmrotate/models/builder.py", line 55, in build_detector
return ROTATED_DETECTORS.build(
File "/opt/conda/envs/POBBv2/lib/python3.8/site-packages/mmcv/utils/registry.py", line 237, in build
return self.build_func(*args, **kwargs, registry=self)
File "/opt/conda/envs/POBBv2/lib/python3.8/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/POBBv2/lib/python3.8/site-packages/mmcv/utils/registry.py", line 72, in build_from_cfg
raise type(e)(f'{obj_cls.name}: {e}')
KeyError: 'RotatedFCOS: "CPMHead: \'vis_train_duration\'"'
(POBBv2)