[Bug] When I tried using RMosaic data augmentation and set the workflow to [('train', 1), ('val', 1)], I encountered an error that showed 'KeyError: img'. However, when I changed the workflow back to [('train', 1)], the error disappeared. Why is this happening?

[wu325]

Prerequisite

• I have searched Issues and Discussions but cannot get the expected help.
• I have read the FAQ documentation but cannot get the expected help.
• The bug has not been fixed in the latest version (master) or latest version (1.x).

Task

I have modified the scripts/configs, or I'm working on my own tasks/models/datasets.

Branch

master branch https://github.com/open-mmlab/mmrotate

Environment

fatal: not a git repository (or any of the parent directories): .git
sys.platform: win32
Python: 3.8.19 (default, Mar 20 2024, 19:55:45) [MSC v.1916 64 bit (AMD64)]
CUDA available: True
GPU 0: NVIDIA GeForce RTX 3060 Ti
CUDA_HOME: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
NVCC: Cuda compilation tools, release 11.8, V11.8.89
MSVC: 用于 x64 的 Microsoft (R) C/C++ 优化编译器 19.40.33811 版
GCC: n/a
PyTorch: 1.8.0+cu111
PyTorch compiling details: PyTorch built with:

• C++ Version: 199711
• MSVC 192829337
• Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications
• Intel(R) MKL-DNN v1.7.0 (Git Hash 7aed236906b1f7a05c0917e5257a1af05e9ff683)
• OpenMP 2019
• CPU capability usage: AVX2
• CUDA Runtime 11.1
• NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
• CuDNN 8.0.5
• Magma 2.5.4
• Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.1, CUDNN_VERSION=8.0.5, CXX_COMPILER=C:/w/b/windows/tmp_bin/sccache-cl.exe, CXX_FLAGS=/DWIN32 /D_WINDOWS /GR /EHsc /w /bigobj -DUSE_PTHREADPOOL -openmp:exper
  imental -DNDEBUG -DUSE_FBGEMM -DUSE_XNNPACK, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.8.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=OFF, USE_NNPACK=OFF, USE_OPENMP=ON,

TorchVision: 0.9.0+cu111
OpenCV: 4.10.0
MMCV: 1.6.0
MMCV Compiler: MSVC 192930137
MMCV CUDA Compiler: 11.1
MMRotate: 0.3.4+

Reproduces the problem - code sample

新配置继承基础模型的设置

from mmdet.core.evaluation import class_names

base = '../../configs/lsknet/lsk_s_fpn_1x_dota_le90.py'

1.数据集设置

dataset_type = 'DOTADataset'
data_root = 'data/test/'
angle_version = 'le90'
classes = ('dam', 'groyne', 'lock', 'sluice', 'weir')
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='RRandomFlip',
flip_ratio=[0.25, 0.25, 0.25],
direction=['horizontal', 'vertical', 'diagonal'],
version=angle_version),
dict(
type='RMosaic', # 马赛克的数据增强
img_scale=(512, 512),
center_ratio_range=(0.8, 1.2),
min_bbox_size=6,
bbox_clip_border=True,
skip_filter=True,
version=angle_version),
dict(
type='PolyRandomRotate', # 对图像和边界框（bbox）进行旋转的数据增强
rotate_ratio=0.5,
angles_range=180,
auto_bound=False,
version=angle_version),
dict(
type='RandomPhotoMetricDistortion', # 光度变换的数据增强
prob=0.5,
brightness_delta=0, # 亮度变化范围
contrast_range=(0.9, 1.1), # 对比度变化范围
saturation_range=(0.9, 1.1), # 饱和度变化范围
hue_delta=18), # 色调变化范围
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(
samples_per_gpu=2,
workers_per_gpu=4,
train=dict(
delete=True,
type='MultiImageMixDataset',
dataset=dict(
type=dataset_type,
classes=classes,
ann_file=data_root + 'train/annfiles/',
img_prefix=data_root + 'train/images/',
pipeline=[
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True)
],
version=angle_version
),
pipeline=train_pipeline,
max_refetch=500),
val=dict(
type=dataset_type,
classes=classes,
ann_file=data_root + 'val/annfiles/',
img_prefix=data_root + 'val/images/'),
test=dict(
type=dataset_type,
classes=classes,
ann_file=data_root + 'test/annfiles/',
img_prefix=data_root + 'test/images/'))

2. 优化器设置

optimizer = dict(
delete=True,
type='AdamW',
lr=0.0001, #/8*gpu_number,
betas=(0.9, 0.999),
weight_decay=0.05)
lr_config = dict(
delete=True,
policy='CosineAnnealing',
by_epoch=False,
min_lr=0,
warmup='linear',
warmup_iters=500,
warmup_ratio=1.0 / 3)
evaluation = dict(interval=1, metric='mAP', save_best='mAP')
runner = dict(type='EpochBasedRunner', max_epochs=12)

3. 结果输出设置

log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
dict(type='TensorboardLoggerHook')
])
workflow = [('train',1),('val',1)]

4. 模型设置

gpu_number = 1 # 修改为自己的GPU数量
model = dict(
backbone=dict(
init_cfg=dict(type='Pretrained', checkpoint='checkpoints/lsk_s_backbone.pth.tar'),
norm_cfg=dict(type='BN', requires_grad=True)), # 单卡训练时将SyncBN改为BN
roi_head=dict(
bbox_head=dict(
num_classes=5)))

Reproduces the problem - command or script

python tools/train.py my_demo/configs/my_lsk_s_orcnn_fpn_1x_dota_le90.py --work-dir my_demo/output/debug/lsk_orcnn

Reproduces the problem - error message

2024-09-04 11:10:24,555 - mmrotate - INFO - workflow: [('train', 1), ('val', 1)], max: 12 epochs
2024-09-04 11:10:24,555 - mmrotate - INFO - Checkpoints will be saved to G:\mmrotate\my_demo\output\debug\lsk_orcnn by HardDiskBackend.
D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmdet\models\dense_heads\anchor_head.py:123: UserWarning: DeprecationWarning: anchor_generator is deprecated, please use "prior_generator" instead
warnings.warn('DeprecationWarning: anchor_generator is deprecated, '
2024-09-04 11:11:20,375 - mmrotate - INFO - Epoch [1][50/95] lr: 3.969e-05, eta: 0:20:14, time: 1.114, data_time: 0.271, memory: 7126, loss_rpn_cls: 0.5340, loss_rpn_bbox: 0.1305, loss_cls: 0.3892, acc: 92.8320, loss_bbox: 0.0319, loss: 1.0856, grad_norm: 12.6486
2024-09-04 11:11:51,915 - mmrotate - INFO - Saving checkpoint at 1 epochs
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 7/7, 1.0 task/s, elapsed: 7s, ETA: 0s2024-09-04 11:12:11,302 - mmrotate - INFO -
+--------+-----+------+--------+-------+
| class | gts | dets | recall | ap |
+--------+-----+------+--------+-------+
| dam | 2 | 129 | 0.000 | 0.000 |
| groyne | 0 | 6 | 0.000 | 0.000 |
| lock | 0 | 3 | 0.000 | 0.000 |
| sluice | 5 | 8 | 0.000 | 0.000 |
| weir | 0 | 15 | 0.000 | 0.000 |
+--------+-----+------+--------+-------+
| mAP | | | | 0.000 |
+--------+-----+------+--------+-------+
2024-09-04 11:12:11,340 - mmrotate - INFO - Exp name: my_lsk_s_orcnn_fpn_1x_dota_le90.py
2024-09-04 11:12:11,340 - mmrotate - INFO - Epoch(val) [1][7] mAP: 0.0000
Traceback (most recent call last):
File "tools/train.py", line 192, in
main()
File "tools/train.py", line 181, in main
train_detector(
File "g:\mmrotate\mmrotate\apis\train.py", line 141, in train_detector
runner.run(data_loaders, cfg.workflow)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmcv\runner\epoch_based_runner.py", line 136, in run
epoch_runner(data_loaders[i], **kwargs)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\autograd\grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmcv\runner\epoch_based_runner.py", line 68, in val
for i, data_batch in enumerate(self.data_loader):
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\utils\data\dataloader.py", line 517, in next
data = self._next_data()
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\utils\data\dataloader.py", line 1199, in _next_data
return self._process_data(data)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\utils\data\dataloader.py", line 1225, in _process_data
data.reraise()
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch_utils.py", line 429, in reraise
raise self.exc_type(msg)
KeyError: Caught KeyError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\utils\data_utils\worker.py", line 202, in _worker_loop
data = fetcher.fetch(index)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\utils\data_utils\fetch.py", line 44, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\torch\utils\data_utils\fetch.py", line 44, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmdet\datasets\custom.py", line 218, in getitem
data = self.prepare_train_img(idx)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmdet\datasets\custom.py", line 241, in prepare_train_img
return self.pipeline(results)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmdet\datasets\pipelines\compose.py", line 41, in call
data = t(data)
File "D:\Program\Anaconda3\envs\mmrotate\lib\site-packages\mmdet\datasets\pipelines\transforms.py", line 469, in call
results[key], direction=results['flip_direction'])
KeyError: 'img'

Additional information

This is just a test dataset, so the issue of mAP=0 can be ignored.