keeping empty tensor for fgradInput in ConvMM - Githubissues

torch / demos

Demos and tutorials around Torch7.

355 stars 301 forks source link

keeping empty tensor for fgradInput in ConvMM #18

Closed jhjin closed 9 years ago

jhjin commented 9 years ago

nilling the fgradInput in ConvolutionMM causes failure in post-training