Week 13 & 14 - Training CNN for raw samples

less than 1 minute read

Proposed net

I have defined an CNN network to address prediction with raw images and non-recurrent networks. The defined network is as follows:

CNN Net

Linear dataset results

  • 1 DOF, 8000 training samples, 1000 test samples
CNN-Linear 1 DOF
  • 2 DOF, 80000 training samples, 10000 test samples
CNN-Linear 2 DOF

Parabolic dataset results

  • 1 DOF, 80000 training samples, 10000 test samples
CNN-Parabolic 1 DOF
  • 2 DOF, 80000 training samples, 10000 test samples
CNN-Parabolic 2 DOF
  • 3 DOF, 80000 training samples, 10000 test samples
CNN-Parabolic 3 DOF

Sinusoidal dataset results

  • 1 DOF, 80000 training samples, 10000 test samples
CNN-Sinusoidal 1 DOF
  • 2 DOF, 80000 training samples, 10000 test samples
CNN-Sinusoidal 2 DOF
  • 3 DOF, 80000 training samples, 10000 test samples
CNN-Sinusoidal 3 DOF

Conclusions

We have obtained good prediction results for linear motio in all its DOF, for the parabolic we obtain the limit in the most complex case (3 DOF), and for the sinusoidal it remains at 2 DOF. This shows that it is more difficult for the network to process raw images than modeled ones since the number of values in each sample is higher in the raw ones.

You May Also Enjoy