mmclassification tutorial colab

mmclassification/docs/en/getting_started.md

데이터셋 준비

https://mmclassification.readthedocs.io/en/latest/getting_started.html#prepare-datasets

...
dataset_type = 'MCSPilot'
CLASSES = ['Normal','Alarm']
DATA_ROOT='/raid/templates/msc_pilot/images' #'data/imagenet',
...

~/my/git/msc/ mmclassification/mmcls/datasets/init.py 에 등록

...

__all__ = [
'BaseDataset', 'ImageNet', 'CIFAR10', 'CIFAR100', 'MNIST', ..... 'CustomDataset',
'MCSPilot'
]

dataset (middleformat을 위한 def load_annotations())

~/my/git/msc/ mmclassification/mmcls/datasets/sixx/mydataset.py

~/my/git/msc/ mmclassification/mmcls/datasets/sixx/mscpilot.py

(openmmlab)$ pwd
/home/oschung_skcc/my/git/msc/mmclassification/configs/swin_transformer

(openmmlab)$ cp ./swin_small_224_b16x64_300e_imagenet.py ../sixx

config 준비

configs/swin_transformer/

(openmmlab)$ pwd
/home/oschung_skcc/my/git/msc/mmclassification/configs/swin_transformer

(openmmlab)$ cp ./swin_small_224_b16x64_300e_imagenet.py ../sixx

base를 보면서…

base = ‘swin-small_16xb64_in1k.py’

/home/htkim2_skcc/git/mmclassification/configs/swin_transformer/swin-small_16xb64_in1k.py

model, datasets, schedules(pipeline), runtime

base = [
‘../base/models/swin_transformer/small_224.py’,
‘../base/datasets/imagenet_bs64_swin_224.py’,
‘../base/schedules/imagenet_bs1024_adamw_swin.py’,
‘../base/default_runtime.py’
]

데모

mmclassification/docs/en/tutorials/MMClassification_python.ipynb

https://colab.research.google.com/github/open-mmlab/mmclassification/blob/master/docs/en/tutorials/MMClassification_python.ipynb

inference

python tools/train.py configs/swin_transformer/kht_swin_small_224_b16x64_300e_imagenet.py