virtual_sketching icon indicating copy to clipboard operation
virtual_sketching copied to clipboard

请问是否解决了呢?我也遇到了相同的问题。

Open yonglezhang95 opened this issue 1 year ago • 6 comments

          请问是否解决了呢?我也遇到了相同的问题。

muten_0_pred

Originally posted by @panda7777777 in https://github.com/MarkMoHR/virtual_sketching/issues/11#issuecomment-1846701715

I operate step by step following readme to test rough_sketches, but I encounter the same question with all simplication sketches is blank white. I can find the what reason leading to this.

yonglezhang95 avatar Aug 05 '24 11:08 yonglezhang95

You could try double checking if the pre-trained model weights are placed and loaded correctly.

MarkMoHR avatar Aug 05 '24 11:08 MarkMoHR

我记得我当时是这样解决的:1,使用CPU推理而非GPU(在我的服务器上,GPU推理并不会更快,而且当时有一些奇怪的警告信息);2,降低 matplotlib 版本。

panda7777777 avatar Aug 05 '24 13:08 panda7777777

另外训练一直是正常的,我可以使用GPU在自己的数据集上训练,并且取得了很好的效果

panda7777777 avatar Aug 05 '24 13:08 panda7777777

我记得我当时是这样解决的:1,使用CPU推理而非GPU(在我的服务器上,GPU推理并不会更快,而且当时有一些奇怪的警告信息);2,降低 matplotlib 版本。

十分感谢,我也用CPU推理就能成功测试出结果了。但是好奇,为啥GPU推理结果就是空白的,我认为应该不是作者说的预训练模型加载不正确的原因。

yonglezhang95 avatar Aug 06 '24 03:08 yonglezhang95

You could try double checking if the pre-trained model weights are placed and loaded correctly.

I have tested successfully via CPU inference, Thank panda 7777777. However, once I change to GPU inference only, the testing results are also blank. I do think that pre-trained model weights are not loaded correctly.

yonglezhang95 avatar Aug 06 '24 03:08 yonglezhang95

那可能是显卡的问题。我那时候是用1080Ti跑的,可能Tensorflow1.x的代码跟现在新的显卡不兼容,可以尝试用旧一点的显卡

MarkMoHR avatar Aug 06 '24 07:08 MarkMoHR