datnt88 / deep-goggle

Source code for "Understanding Deep Image Representations by Inverting Them", CVPR 2015

Directory Structure

.
+-- core
|   +-- invert_nn.m - The core optimization lies here
+-- helpers - Several auxilary functions that may be useful in general
+-- experiments - All the code to replicate our experiments
|   +-- networks
|   |   +-- hog_net.m - The hog and hogb networks are created using this
|   |   +-- dsift_net.m - The dense sift neural network is here
|   |   +-- Other networks used in our experiments can be downloaded from http://www.robots.ox.ac.uk/~aravindh/networks.html
+-- ihog - either copy or soft link ihog from Vondrick et. al. This is required to run our experiments with hoggle.
+-- matconvnet - either copy or soft link matconvnet code here. If this is not here, then the setup function will not work.
+-- vlfeat - again either copy or soft copy. If this is not here, then the setup function will not work.

Experiments from the paper

To run the experiments used for our publication and replicate their results please follow the instructions below

Get the images Download/soft link the imagenet validation images into experiments/data/imagenet12-val Download/soft link the stock abstrack images into experiments/data/stock

For any of the cases below you need to run the following in matlab

cd experiments;
experiment_setup;

I) Experiment for a single reconstruction across all layers of

experiment_cnn;

See the results in data/results/ #TODO - List of the files relevant here TODO - Add experiment_xxx files here as and when they are documented.

Setting up and running your own networks

1. Create a network (net) that is compatible with matconvnet vl_simplenn function.
2. Run dg_setup.m in matlab
3. Run the network forward to generate a target reference representation y0
4. Call res = invert_nn(net, y0, [options]);
5. res.output{end} is the required reconstruction.