自定义autograd function

在TSN代码中， segmentconsensus是一个自定义函数，所以要写一下它对应的梯度运算


# tj : https://blog.csdn.net/tsq292978891/article/details/79364140
class SegmentConsensus(torch.autograd.Function):@staticmethoddef forward(ctx, input_tensor, consensus_type, dim):ctx.consensus_type = consensus_typectx.dim = dimctx.shape = input_tensor.size()if consensus_type == 'avg':output = input_tensor.mean(dim=dim, keepdim=True)elif consensus_type == 'identity':output = input_tensorelse:output = Nonereturn output@staticmethoddef backward(ctx, grad_output):if ctx.consensus_type == 'avg':grad_in = grad_output.expand(ctx.shape) / float(ctx.shape[ctx.dim])elif ctx.consensus_type == 'identity':grad_in = grad_outputelse:grad_in = Nonereturn grad_in, None, None# tj : 常数要返回 None


# tj : https://blog.csdn.net/tsq292978891/article/details/79364140
class ConsensusModule(torch.nn.Module):def __init__(self, consensus_type, dim=1):super(ConsensusModule, self).__init__()self.consensus_type = consensus_type if consensus_type != 'rnn' else 'identity'self.dim = dimdef forward(self, input):return SegmentConsensus.apply(input, self.consensus_type, self.dim)

自定义autograd function相关推荐

Pytorch使用autograd.Function自定义拓展神经网络
我们知道CNN这类人工神经网络都基于BP算法进行优化,因此需要误差关于权重是连续可导的,这是可以运用BP算法的前提条件:也有一些网络不满足这个条件. 1.可导对于可连续求导的神经网络构建时采用nn. ...
Pytorch的自定义拓展:torch.nn.Module和torch.autograd.Function
参考链接:pytorch的自定义拓展之(一)--torch.nn.Module和torch.autograd.Function_LoveMIss-Y的博客-CSDN博客_pytorch自定义backw ...
使用torch.autograd.function解决dist.all_gather不能反向传播问题
1. 问题来源最近在用mmcv复现Partial FC模型,看到源码中,有单独写的前向反向传播,甚是疑惑- 源码: # Features all-gather total_features = to ...
Please use new-style autograd function with static forward method
Please use new-style autograd function with static forward method Legacy autograd function with non- ...
Legacy autograd function with non-static forward method is deprecated. Please use new-style autograd
Legacy autograd function with non-static forward method is deprecated. Please use new-style autograd ...
数据库系列之mysql 自定义函数function，函数和存储过程的区别
mysql 自定义函数function,函数和存储过程的区别 https://blog.csdn.net/u010365819/article/details/80470448 https://blo ...
Pytorch版本过高产生的RuntimeError: Legacy autograd function with non-static forward method is deprecated.
前言在尝试用ECO Lite做视频分类的时候,使用了作者的Pytorch实现,然而Pytorch实现是基于Pytorch0.4的,我自己的Pytorch版本是1.4,所以在跑模型的时候出现了一些问题 ...
Legacy autograd function with non-static forward method is deprecated
Legacy autograd function with non-static forward method is deprecated 在网络s3fd_atss_sapd 测试时发现这个问题, d ...
RuntimeError: Legacy autograd function with non-static forward method is deprecated.
显卡RTX3080 cuda 11.1 cudnn 8.0.5 python3.6.4 在使用pytorch1.0训练densefusion模型时报错,改用pytorch1.7,然后报上面的错误 Tr ...

自定义autograd function

自定义autograd function相关推荐

最新文章

热门文章