Texture Synthesis Using Convolutional Neural Networks
Authors: Leon Gatys, Alexander S. Ecker, Matthias Bethge
NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Samples from the model are of high perceptual quality demonstrating the generative power of neural networks trained in a purely discriminative fashion. ... We show that across layers the texture representations increasingly capture the statistical properties of natural images while making object information more and more explicit. ... We show textures generated by our model from four different source images (Fig. 2). |
| Researcher Affiliation | Academia | Leon A. Gatys Centre for Integrative Neuroscience, University of T ubingen, Germany Bernstein Center for Computational Neuroscience, T ubingen, Germany Graduate School of Neural Information Processing, University of T ubingen, Germany leon.gatys@bethgelab.org Alexander S. Ecker Centre for Integrative Neuroscience, University of T ubingen, Germany Bernstein Center for Computational Neuroscience, T ubingen, Germany Max Planck Institute for Biological Cybernetics, T ubingen, Germany Baylor College of Medicine, Houston, TX, USA Matthias Bethge Centre for Integrative Neuroscience, University of T ubingen, Germany Bernstein Center for Computational Neuroscience, T ubingen, Germany Max Planck Institute for Biological Cybernetics, T ubingen, Germany |
| Pseudocode | No | The paper describes procedures and equations but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | Yes | Source code to generate textures with CNNs as well as the rescaled VGG-19 network can be found at http://github.com/leongatys/Deep Textures |
| Open Datasets | Yes | For each layer we computed the Gram-matrix representation of each image in the Image Net training set [23] and trained a linear soft-max classifier to predict object identity. ... [23] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei. Image Net Large Scale Visual Recognition Challenge. ar Xiv:1409.0575 [cs], September 2014. ar Xiv: 1409.0575. |
| Dataset Splits | Yes | We computed the accuracy of these linear classifiers on the Image Net validation set and compared them to the performance of the original VGG-19 network also evaluated on the 224 224 centre crops of the validation images. |
| Hardware Specification | No | The paper mentions 'using GPUs' but does not provide specific models or any other detailed hardware specifications. |
| Software Dependencies | No | The paper mentions 'caffe-framework [12]' and 'L-BFGS [30]' but does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | For texture generation we found that replacing the maxpooling operation by average pooling improved the gradient flow and one obtains slightly cleaner results, which is why the images shown below were generated with average pooling. Finally, for practical reasons, we rescaled the weights in the network such that the mean activation of each filter over images and positions is equal to one. ... In our work we use L-BFGS [30] ... for the loss terms above a certain layer we set the weights wl = 0, while for the loss terms below and including that layer, we set wl = 1. |