How can I train a subset of a custom data set

[wangzhaoyang-508]

I successfully registered a custom dataset with 8 classes of objects. And it can be trained well.

Now, 4 of the 8 classes in my custom dataset is no need to be detected anymore。

I don't want to change the json file, so how do I re-register， so that the model can only focuses on the four useful objects when it trained?

in detrex I tried to use the “MetadataCatalog.get” but it can not help the error is AssertionError: Attribute 'thing_classes' in the metadata of 'my_eldataset_train' cannot be set to a different value! ['duanshan', 'heiban', 'xianzhuangquexian', 'xuhan'] != ['duanshan', 'heiban', 'xianzhuangquexian', 'xuhan', 'yinlie', 'crush', 'finger', 'star']

codes import itertools from omegaconf import OmegaConf

import detectron2.data.transforms as T from detectron2.config import LazyCall as L from detectron2.data import ( build_detection_test_loader, build_detection_train_loader, get_detection_dataset_dicts, MetadataCatalog, ) from detectron2.data.datasets import register_coco_instances from detectron2.evaluation import COCOEvaluator

from detrex.data import DetrDatasetMapper

dataloader = OmegaConf.create()

register_coco_instances("my_eldataset_train", {}, '/data1/wzydatasets/yuanle/coco/annotations/instances_train2017.json', '/data1/wzydatasets/yuanle/coco/train2017/') register_coco_instances("my_eldataset_test", {}, '/data1/wzydatasets/yuanle/coco/annotations/instances_test2017.json', '/data1/wzydatasets/yuanle/coco/test2017/')

MetadataCatalog.get("my_eldataset_train").thing_classes = ['duanshan', 'heiban', 'xianzhuangquexian', 'xuhan'] MetadataCatalog.get("my_eldataset_test").thing_classes = ['duanshan', 'heiban', 'xianzhuangquexian', 'xuhan']

dataloader.train = L(build_detection_train_loader)( dataset=L(get_detection_dataset_dicts)(names="my_eldataset_train"), mapper=L(DetrDatasetMapper)( augmentation=[ L(T.ResizeShortestEdge)( short_edge_length=600, max_size=600, ), L(T.RandomFlip)(), L(T.ResizeShortestEdge)( short_edge_length=(320, 480, 512, 544, 576, 608,),

max_size=1333,

max_size=640, sample_style="choice", ), ], augmentation_with_crop=[ L(T.RandomFlip)(), L(T.ResizeShortestEdge)( short_edge_length=600, max_size=600, ), L(T.RandomCrop)( crop_type="absolute_range", crop_size=(300, 400), # 必须是列表，必须有两个参数，否则会报错。 ), L(T.ResizeShortestEdge)(

short_edge_length=(480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800),

short_edge_length=(320, 480, 512, 544, 576, 608,),

max_size=1333,

max_size=640, sample_style="choice", ), ], is_train=True, mask_on=False, img_format="RGB", ), total_batch_size=16, num_workers=4, )

dataloader.test = L(build_detection_test_loader)( dataset=L(get_detection_dataset_dicts)(names="my_eldataset_test", filter_empty=False), mapper=L(DetrDatasetMapper)( augmentation=[ L(T.ResizeShortestEdge)( short_edge_length=600, max_size=640, ), ], augmentation_with_crop=None, is_train=False, mask_on=False, img_format="RGB", ), num_workers=4, )

dataloader.evaluator = L(COCOEvaluator)( dataset_name="${..test.dataset.names}", )

[bouachalazhar]

I would like to do it with 17 classes from COCO DATASET.