search

Evolving-AI-Lab / mfv

Multifaceted Feature Visualization
70 stars 22 forks source link

Visualizations in Convolutional Layers #2

Open Lucaszw opened 8 years ago

Lucaszw commented 8 years ago

Great Work!

In figure 6 from the ICML workshop paper you showed visualizations for convolution layers as well as fc layers. Any suggestions on how to replicate this in the sample code?

Thanks!

anguyen8 commented 8 years ago

@Lucaszw : The code works for both fc and conv units, i.e. you could use these parameters: https://github.com/Evolving-AI-Lab/mfv/blob/master/act_max.tvd.mean.py#L235-L238

To replicate Fig. 6, you'd need to do all the steps of 1) compute the mean images from the validation set, 2) optimize images starting from those mean images (via the current code in this repo). The code for (1) is straightforward (as described in the paper) but I haven't had time yet to polish and publish here.

Let me know if you have other questions!