KeyError: 'SemiBalanceSampler is not in the sampler registry'

[poorguy2001]

Traceback (most recent call last): File "tools/train_detr_ssod.py", line 215, in main() File "tools/train_detr_ssod.py", line 202, in main train_detector( File "/home/lg/code/Semi-DETR-main/detr_ssod/apis/train.py", line 69, in train_detector data_loaders = [ File "/home/lg/code/Semi-DETR-main/detr_ssod/apis/train.py", line 70, in build_dataloader( File "/home/lg/code/Semi-DETR-main/detr_ssod/datasets/builder.py", line 69, in build_dataloader build_sampler(sampler_cfg, default_args=default_sampler_cfg) File "/home/lg/code/Semi-DETR-main/detr_ssod/datasets/builder.py", line 41, in build_sampler return build_from_cfg(cfg, SAMPLERS, default_args) File "/home/lg/.conda/envs/semidetr/lib/python3.8/site-packages/mmcv/utils/registry.py", line 44, in build_from_cfg raise KeyError( KeyError: 'SemiBalanceSampler is not in the sampler registry' 在训练train_detr_ssod.py时，使用detr_ssod_dino_detr_r50_coco_120k.py，发现KeyError: 'SemiBalanceSampler is not in the sampler registry'。经过检查，在base_dino_detr_ssod_coco.py的第223行定义了sampler=dict( train=dict( type="SemiBalanceSampler", sample_ratio=[1, 4], by_prob=True,

at_least_one=True,

        epoch_length=7330,
    )但是在detr_ssod/datasets/__init__.py并没有SemiBalanceSampler，只有DistributedGroupSemiBalanceSampler，请问这是代码的问题还是什么？

[poorguy2001]

我大概明白了，作者的代码是只支持多个GPU进行分布式运行，但对于单个GPU应该怎么办呢？？

[JCZ404]

Thanks for your interest, you can specify the GPU number to 1 when launching the distributed training script to enable single GPU running but still allow the DistributedGroupSemiBalanceSampler, like sh tools/dist_train_detr_ssod.sh dino_detr_ssod 1 10 1. But running with a single GPU is recommended only when you want to debug the code. If you want to reproduce our performance, you should run with 8 GPU cards following the experiment setting in our paper.