Converting 4D matrix to 2D

Sivaramakrishnan Rajaraman

2017 11 28

0 个回答

1 次查看（30 天）

0 个投票

I'm extracting features from pre-trained CNNs. The size of the extracted feature matrix is in 4-D. The extracted features are then used to train a linear classifier. The extracted features have dimensions (1,1,1024,200) (in general m,n,p,q) where 1024 is the dimension of the features and 200 is the number of images. I'd like to extract only the p and q components,leaving behind the first and second dimensions. I tried to use Squeeze function but it doesn't help. Kindly help me with the syntax to obtain only the 2-dimensions (i.e.)(1024,200).

2 个评论
显示无隐藏无

Matt J 2017-11-28

Try the squeeze function again. There's no reason it shouldn't have worked.

Sivaramakrishnan Rajaraman 2017-11-28

The squeeze function works only at the 'avg_pool' layer, just before the softmax. It doesn't show any error when applied to the earlier layers, however, the dimensions remain unchanged. ResNet50 is a DAG network. Does that have anything to do with restricting the usage of Squeeze in the earlier layers?

请先登录，再进行评论。

请先登录，再回答此问题。

Follow Question