KeyError: 'ActionVisualizer is not in the mmengine::visualizer registry.[Bug]
berengueradrian opened this issue · comments
Branch
main branch (1.x version, such as v1.0.0, or dev-1.x branch)
Prerequisite
- I have searched Issues and Discussions but cannot get the expected help.
- I have read the documentation but cannot get the expected help.
- The bug has not been fixed in the latest version.
Environment
I am trying to fine tune my own dataset in AVA format for Spatio Temporal Action Detection in a Docker environment:
ARG PYTORCH="1.13.1"
ARG CUDA="11.6"
ARG CUDNN="8"
FROM pytorch/pytorch:${PYTORCH}-cuda${CUDA}-cudnn${CUDNN}-devel
ENV TORCH_CUDA_ARCH_LIST="6.0 6.1 7.0+PTX"
ENV TORCH_NVCC_FLAGS="-Xfatbin -compress-all"
ENV CMAKE_PREFIX_PATH="$(dirname $(which conda))/../"
# fetch the key refer to https://forums.developer.nvidia.com/t/18-04-cuda-docker-image-is-broken/212892/9
RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub 32
RUN apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/7fa2af80.pub
RUN apt-get update && apt-get install -y git ninja-build libglib2.0-0 libsm6 libxrender-dev libxext6 ffmpeg \
&& apt-get clean \
&& rm -rf /var/lib/apt/lists/*
# Install MMCV
RUN pip install wheel
RUN pip install openmim
RUN mim install mmengine mmcv==2.0.0rc4
Then i installed the following:
pip install cython --no-cache-dir
pip install --no-cache-dir -e .
pip install pytorchvideo
Describe the bug
This is the complete log of the bug:
05/07 08:32:33 - mmengine - WARNING - Failed to import `None.registry` make sure the registry.py exists in `None` package.
05/07 08:32:33 - mmengine - WARNING - Failed to search registry with scope "mmdet" in the "visualizer" registry tree. As a workaround, the current "visualizer" registry in "mmengine" is used to build instance. This may cause unexpected failure when running the built modules. Please check whether "mmdet" is a correct scope, or whether the registry is initialized.
Traceback (most recent call last):
File "/workspace/mmaction2/tools/train.py", line 143, in <module>
main()
File "/workspace/mmaction2/tools/train.py", line 132, in main
runner = Runner.from_cfg(cfg)
File "/opt/conda/lib/python3.10/site-packages/mmengine/runner/runner.py", line 462, in from_cfg
runner = cls(
File "/opt/conda/lib/python3.10/site-packages/mmengine/runner/runner.py", line 416, in __init__
self.visualizer = self.build_visualizer(visualizer)
File "/opt/conda/lib/python3.10/site-packages/mmengine/runner/runner.py", line 803, in build_visualizer
return VISUALIZERS.build(visualizer)
File "/opt/conda/lib/python3.10/site-packages/mmengine/registry/registry.py", line 570, in build
return self.build_func(cfg, *args, **kwargs, registry=self)
File "/opt/conda/lib/python3.10/site-packages/mmengine/registry/build_functions.py", line 100, in build_from_cfg
raise KeyError(
KeyError: 'ActionVisualizer is not in the mmengine::visualizer registry. Please check whether the value of `ActionVisualizer` is correct or it was registered as expected. More details can be found at https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#import-the-custom-module'
Reproduces the problem - code sample
This is the code I have in my configs/detection/slowfast folder:
_base_ = [
'../../_base_/models/slowfast_r50.py',
'../../_base_/default_runtime.py'
]
default_scope = "mmdet"
url = ('https://download.openmmlab.com/mmaction/recognition/slowfast/'
'slowfast_r50_8x8x1_256e_kinetics400_rgb/'
'slowfast_r50_8x8x1_256e_kinetics400_rgb_20200716-73547d2b.pth')
model = dict(
type='FastRCNN',
_scope_='mmdet',
init_cfg=dict(type='Pretrained', checkpoint=url),
backbone=dict(
type='mmaction.ResNet3dSlowFast',
resample_rate=4,
speed_ratio=4,
channel_ratio=8,
pretrained=None,
slow_pathway=dict(
type='resnet3d',
depth=50,
pretrained=None,
lateral=True,
conv1_kernel=(1, 7, 7),
dilations=(1, 1, 1, 1),
conv1_stride_t=1,
pool1_stride_t=1,
inflate=(0, 0, 1, 1),
spatial_strides=(1, 2, 2, 1),
fusion_kernel=7),
fast_pathway=dict(
type='resnet3d',
depth=50,
pretrained=None,
lateral=False,
base_channels=8,
conv1_kernel=(5, 7, 7),
conv1_stride_t=1,
pool1_stride_t=1,
spatial_strides=(1, 2, 2, 1))),
roi_head=dict(
type='AVARoIHead',
bbox_roi_extractor=dict(
type='SingleRoIExtractor3D',
roi_layer_type='RoIAlign',
output_size=8,
with_temporal_pool=True),
bbox_head=dict(
type='BBoxHeadAVA',
background_class=True,
in_channels=2304,
num_classes=8,
multilabel=False,
dropout_ratio=0.5)),
data_preprocessor=dict(
type='mmaction.ActionDataPreprocessor',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
format_shape='NCTHW'),
train_cfg=dict(
rcnn=dict(
assigner=dict(
type='MaxIoUAssignerAVA',
pos_iou_thr=0.9,
neg_iou_thr=0.9,
min_pos_iou=0.9),
sampler=dict(
type='RandomSampler',
num=16,
pos_fraction=1,
neg_pos_ub=-1,
add_gt_as_proposals=True),
pos_weight=1.0)),
test_cfg=dict(rcnn=None))
data_root = './data/data_ava-kinetics/train'
data_root_val = './data/data_ava-kinetics/val'
ann_file_train = './data/data_ava-kinetics/train.txt'
ann_file_val = './data/data_ava-kinetics/val.txt'
ann_file_test = './data/data_ava-kinetics/test.txt'
dataset_type = 'AVADataset'
data_root = './data/data_ava-kinetics/train'
anno_root = './data/data_ava-kinetics'
ann_file_train = f'{anno_root}/train.txt'
ann_file_val = f'{anno_root}/val.txt'
exclude_file_train = None
exclude_file_val = None
label_file = f'{anno_root}/label_file.pbtxt'
proposal_file_train = None
proposal_file_val = None
file_client_args = dict(io_backend='disk')
train_pipeline = [
dict(type='SampleAVAFrames', clip_len=32, frame_interval=2),
dict(type='RawFrameDecode', **file_client_args),
dict(type='RandomRescale', scale_range=(256, 320)),
dict(type='RandomCrop', size=256),
dict(type='Flip', flip_ratio=0.5),
dict(type='FormatShape', input_format='NCTHW', collapse=True),
dict(type='PackActionInputs')
]
# The testing is w/o. any cropping / flipping
val_pipeline = [
dict(
type='SampleAVAFrames', clip_len=32, frame_interval=2, test_mode=True),
dict(type='RawFrameDecode', **file_client_args),
dict(type='Resize', scale=(-1, 256)),
dict(type='FormatShape', input_format='NCTHW', collapse=True),
dict(type='PackActionInputs')
]
train_dataloader = dict(
batch_size=8,
num_workers=8,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=True),
dataset=dict(
type=dataset_type,
ann_file=ann_file_train,
exclude_file=exclude_file_train,
pipeline=train_pipeline,
label_file=label_file,
proposal_file=proposal_file_train,
data_prefix=dict(img=data_root)))
val_dataloader = dict(
batch_size=1,
num_workers=8,
persistent_workers=True,
sampler=dict(type='DefaultSampler', shuffle=False),
dataset=dict(
type=dataset_type,
ann_file=ann_file_val,
exclude_file=exclude_file_val,
pipeline=val_pipeline,
label_file=label_file,
proposal_file=proposal_file_val,
data_prefix=dict(img=data_root),
test_mode=True))
test_dataloader = val_dataloader
val_evaluator = dict(
type='AVAMetric',
ann_file=ann_file_val,
label_file=label_file,
exclude_file=exclude_file_val)
test_evaluator = val_evaluator
train_cfg = dict(
type='EpochBasedTrainLoop', max_epochs=20, val_begin=1, val_interval=1)
val_cfg = dict(type='ValLoop')
test_cfg = dict(type='TestLoop')
param_scheduler = [
dict(type='LinearLR', start_factor=0.1, by_epoch=True, begin=0, end=5),
dict(
type='MultiStepLR',
begin=0,
end=20,
by_epoch=True,
milestones=[10, 15],
gamma=0.1)
]
optim_wrapper = dict(
optimizer=dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.00001),
clip_grad=dict(max_norm=40, norm_type=2))
# Default setting for scaling LR automatically
# - `enable` means enable scaling LR automatically
# or not by default.
# - `base_batch_size` = (8 GPUs) x (8 samples per GPU).
auto_scale_lr = dict(enable=False, base_batch_size=64)
Reproduces the problem - command or script
I run the training like this:
python tools/train.py configs/detection/slowfast/slowfast_kinetics400-pretrained-r50_8xb8-8x8x1-20e_chantwin.py
Reproduces the problem - error message
No response
Additional information
- I am trying to obtain results of action detection from my own dataset which contains different action classes, is not multilabel and is not annotated by key frames.
- My dataset is composed of about 170 videos ranging from 20s to 1m30s of duration. Each video will contain 1 or more bboxes of the same class but not different classes. Also, it will contain just 1 activity label per bounding box per frame.