火锅女孩

Results 16 comments of 火锅女孩

Why My CTC Loss is always Nan or Inf value?

I'm in the same situation. Have you solved it?

The generated mask graph is in the shape of [h, W, 3]. Now, the following data that only supports three-dimensional data appears using celoss. Does it need to convert the three-channel graph into one channel? Finally, do a dimensionality reduction to get the shape of [n, h, w] ?

![image](https://user-images.githubusercontent.com/82756842/140675390-0026077c-b84f-4b32-a14e-9046d7340eef.png)

The generated mask graph is in the shape of [h, W, 3]. Now, the following data that only supports three-dimensional data appears using celoss. Does it need to convert the three-channel graph into one channel? Finally, do a dimensionality reduction to get the shape of [n, h, w] ?

ok,thank U.its good.

训练结果不好，怎么调参数

用了200张图片，跑了200轮，训练集当测试集，效果如下。不知道那个环节出了问题。 ![image](https://user-images.githubusercontent.com/82756842/141034945-51e01630-d98f-437b-af40-3bba7da2f091.png)

训练结果不好，怎么调参数

不知道那个环节出了问题

训练结果不好，怎么调参数

![image](https://user-images.githubusercontent.com/82756842/141043925-2facf20e-7e52-4c71-836b-9fb5ea1c705c.png) The loss is strange，and the learning rate has dropped to 0

请问，网络输入修改为224后，会提升多少呢？

你好，你修改成224后有什么神奇的效果吗？

ImportError: cannot import name 'deform_conv_cuda' from 'assets.ops.dcn'

> 导入错误：无法从“assets.ops.dcn”导入名称“deform_conv_cuda” **大佬，解决了吗？你能告诉我吗？**

Hello, help me. If you want to make this kind of data, what file should be used to save it?

Because the txt file does not support saving this character content

请问一下测试速度很慢，而且中途显存就爆了是为什么呢？训练没有问题，测试集只有500张

请问你解决了吗？我训练resnet50时遇到同样的问题，val时显存一致往上涨，直到爆炸