Meet RuntimeError about scalar type when train faster-rcnn on custom dataset

[DarkstartsUp]

Error report:
2019-04-01 16:58:20,122 - INFO - Start running, host: hostname@hostname1, work_dir: /home/hostname/mmdetection/work_dirs/my_faster_rcnn_r50_fpn_1x
2019-04-01 16:58:20,122 - INFO - workflow: [('train', 1)], max: 30 epochs
Traceback (most recent call last):
File "./tools/train.py", line 90, in
main()
File "./tools/train.py", line 86, in main
logger=logger)
File "/home/hostname/mmdetection/mmdet/apis/train.py", line 57, in train_detector
_dist_train(model, dataset, cfg, validate=validate)
File "/home/hostname/mmdetection/mmdet/apis/train.py", line 96, in _dist_train
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
File "/home/hostname/anaconda3/envs/dl/lib/python3.6/site-packages/mmcv/runner/runner.py", line 355, in run
epoch_runner(data_loaders[i], **kwargs)
File "/home/hostname/anaconda3/envs/dl/lib/python3.6/site-packages/mmcv/runner/runner.py", line 261, in train
self.model, data_batch, train_mode=True, **kwargs)
File "/home/hostname/mmdetection/mmdet/apis/train.py", line 37, in batch_processor
losses = model(**data)
File "/home/hostname/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/hostname/anaconda3/envs/dl/lib/python3.6/site-packages/mmcv/parallel/distributed.py", line 50, in forward
return self.module(*inputs[0], **kwargs[0])
File "/home/hostname/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in call
result = self.forward(*input, **kwargs)
File "/home/hostname/mmdetection/mmdet/models/detectors/base.py", line 80, in forward
return self.forward_train(img, img_meta, **kwargs)
File "/home/hostname/mmdetection/mmdet/models/detectors/two_stage.py", line 98, in forward_train
*rpn_loss_inputs, gt_bboxes_ignore=gt_bboxes_ignore)
File "/home/hostname/mmdetection/mmdet/models/anchor_heads/rpn_head.py", line 51, in loss
gt_bboxes_ignore=gt_bboxes_ignore)
File "/home/hostname/mmdetection/mmdet/models/anchor_heads/anchor_head.py", line 198, in loss
sampling=sampling)
File "/home/hostname/mmdetection/mmdet/core/anchor/anchor_target.py", line 63, in anchor_target
unmap_outputs=unmap_outputs)
File "/home/hostname/mmdetection/mmdet/core/utils/misc.py", line 24, in multi_apply
return tuple(map(list, zip(*map_results)))
File "/home/hostname/mmdetection/mmdet/core/anchor/anchor_target.py", line 116, in anchor_target_single
anchors, gt_bboxes, gt_bboxes_ignore, None, cfg)
File "/home/hostname/mmdetection/mmdet/core/bbox/assign_sampling.py", line 32, in assign_and_sample
gt_labels)
File "/home/hostname/mmdetection/mmdet/core/bbox/assigners/max_iou_assigner.py", line 76, in assign
overlaps = bbox_overlaps(gt_bboxes, bboxes)
File "/home/hostname/mmdetection/mmdet/core/bbox/geometry.py", line 48, in bbox_overlaps
lt = torch.max(bboxes1[:, None, :2], bboxes2[:, :2]) # [rows, cols, 2]
RuntimeError: Expected object of scalar type Double but got scalar type Float for argument #2 'other'

Part of my custom dataset:
[{'filename': 'IMG_12_1.jpg', 'width': 793, 'height': 595, 'ann': {'bboxes': array([[432, 180, 492, 280],
[450, 108, 550, 168],
[540, 108, 640, 168],
[450, 162, 550, 222],
[486, 90, 546, 190],
[486, 180, 546, 280],
[270, 162, 370, 222],
[360, 216, 460, 276],
[450, 216, 550, 276],
[270, 90, 330, 190],
[594, 90, 654, 190],
[450, 432, 550, 492],
[378, 180, 438, 280],
[432, 450, 492, 550],
[180, 162, 280, 222],
[360, 162, 460, 222],
[630, 216, 730, 276],
[450, 324, 550, 384],
[540, 324, 640, 384],
[216, 90, 276, 190],
[324, 90, 384, 190],
[540, 90, 600, 190],
[702, 180, 762, 280],
[180, 108, 280, 168],
[270, 108, 370, 168]], dtype=int32), 'labels': array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0], dtype=int32), 'bboxes_ignore': array([], dtype=int32), 'labels_ignore': array([], dtype=int32)}}, {'filename': 'IMG_9_1.jpg', 'width': 664, 'height': 373, 'ann': {'bboxes': array([[360, 162, 460, 222],
[ 54, 180, 114, 280],
[378, 180, 438, 280],
[486, 180, 546, 280],
[ 0, 162, 100, 222],
[108, 180, 168, 280],
[450, 162, 550, 222],
[ 90, 162, 190, 222],
[450, 216, 550, 276],
[432, 180, 492, 280],
[324, 180, 384, 280],
[ 90, 216, 190, 276],
[540, 180, 600, 280],
[540, 162, 640, 222],
[ 0, 216, 100, 276],
[180, 216, 280, 276],
[540, 216, 640, 276],
[162, 180, 222, 280],
[ 0, 180, 60, 280],
[594, 180, 654, 280],
[360, 216, 460, 276],
[216, 180, 276, 280],
[270, 162, 370, 222],
[270, 216, 370, 276],
[270, 270, 370, 330]], dtype=int32), 'labels': array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0], dtype=int32), 'bboxes_ignore': array([], dtype=int32), 'labels_ignore': array([], dtype=int32)}}]

Part of config file:

dataset settings

dataset_type = 'CustomDataset'
data_root = 'data/CrowdImages/'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
data = dict(
imgs_per_gpu=2,
workers_per_gpu=2,
train=dict(
type=dataset_type,
ann_file=data_root + 'annotations/annos_train.pkl',
img_prefix=data_root + 'trainImgs/',
img_scale=(1000, 600),
img_norm_cfg=img_norm_cfg,
size_divisor=32,
flip_ratio=0.5,
with_mask=False,
with_crowd=False,
with_label=True),
val=dict(
type=dataset_type,
ann_file=data_root + 'annotations/annos_val.pkl',
img_prefix=data_root + 'valImgs/',
img_scale=(1000, 600),
img_norm_cfg=img_norm_cfg,
size_divisor=32,
flip_ratio=0,
with_mask=False,
with_crowd=False,
with_label=True),
test=dict(
type=dataset_type,
ann_file=data_root + 'annotations/annos_test.pkl',
img_prefix=data_root + 'testImgs/',
img_scale=(1000, 600),
img_norm_cfg=img_norm_cfg,
size_divisor=32,
flip_ratio=0,
with_mask=False,
with_label=True,
test_mode=True))

optimizer

optimizer = dict(type='SGD', lr=0.0025, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))

learning policy

'''
lr_config = dict(
policy='step',
warmup='linear',
warmup_iters=500,
warmup_ratio=1.0 / 3,
step=[8, 11])
'''
lr_config = dict(policy='step', step=[3])
checkpoint_config = dict(interval=1)

yapf:disable

log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
# dict(type='TensorboardLoggerHook')
])

yapf:enable

runtime settings

total_epochs = 30
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/my_faster_rcnn_r50_fpn_1x'
load_from = None
resume_from = None
workflow = [('train', 1)]

Could anyone help? THX!

[DarkstartsUp]

This problem has been solved:
In customed dataset, the datatype of anno should be:
'bboxes': np.array(dtype=np.float32),
'labels': np.array(dtype=np.int64),
'bboxes_ignore': np.array(dtype=np.float32),
'labels_ignore': np.array(dtype=np.int64)

[mejihero]

Hi, I have the same issue when training the model. Can I ask how to sort it out?