Implementation from the CVPR 2018 paper "What do Deep Networks Like to See?".
This is a simple proof of concept that uses a fine-tuned autoencoder on ResNet50 to reconstruct input images.
- Download the Torch model from here and store it with the root directory of the repo.
- Call
python plot_ae_reconstruction.py -i PATH
where PATH
is the path to the input image.
3. An output image should be saved in the root directory of the repo with the reconstruction.
For more info, please check out the paper's website.
- Weights for the original SegNet (pre-trained on YFCC100m) are now available here and can be used by
plot_ae_reconstruction.py
. Make sure the path is correctly loaded by modifying the global variableRESNET_PATH
.