RealBasicVSR训练（二）

cv-daily

已于 2022-10-24 18:07:37 修改

阅读量1.4k

点赞数

文章标签：超分辨率重建 1024程序员节

于 2022-10-24 17:39:22 首次发布

本文链接：https://blog.csdn.net/weixin_41012399/article/details/127496642

版权

直接训练第二步微调的模型。

参考：
https://blog.csdn.net/weixin_53610475/article/details/126306184

按步骤配置后报错：
1.num_samples should be a positive integer value, but got num_samples=0
修改
mmediting/mmedit/datasets/builder.py

if dist:
    sampler = DistributedSampler(
        dataset,
        world_size,
        rank,
        shuffle=shuffle,
        samples_per_gpu=samples_per_gpu,
        seed=seed)
    shuffle = False
    batch_size = samples_per_gpu
    num_workers = workers_per_gpu
else:
    sampler = None
    shuffle = False  ####添加上这一句即可
    batch_size = num_gpus * samples_per_gpu
    num_workers = num_gpus * workers_per_gpu

2.报错：return next(self._sampler_iter) StopIteration
解决：数据路径问题，修改路径即可。

3.修改参数cfg.gpus = 2

AssertionError: MMDataParallel only supports single GPU training, if you need to train with multiple GPUs, please use MMDistributedDataParallel instead.

这样只能用1个gpu训练？