Open scottclowe opened 9 years ago
As performed in this paper: http://www.csc.kth.se/~azizpour/papers/ha_cvpr14w.pdf
We want to be able to extract the deep convolutions from the CNN directly instead of the class predictions from the final layer. This will give us a big feature vector we can feed into sklearn along with the CV features.
https://groups.google.com/forum/#!searchin/theano-users/feature$20output/theano-users/7jL1j0_gk1o/8Ba2gOWz6boJ
https://groups.google.com/forum/#!searchin/theano-users/feature$20output/theano-users/JBn1aNVPaAA/it-zDJut9WwJ
https://groups.google.com/forum/#!searchin/theano-users/feature$20output/theano-users/Xhb5KQlU88g/DqDlxvPfsGYJ
scottclowe
commented
9 years ago
As performed in this paper: http://www.csc.kth.se/~azizpour/papers/ha_cvpr14w.pdf
We want to be able to extract the deep convolutions from the CNN directly instead of the class predictions from the final layer. This will give us a big feature vector we can feed into sklearn along with the CV features.