Generating Images with Perceptual Similarity Metrics based on Deep Networks

Authors: Alexey Dosovitskiy, Thomas Brox

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate two examples of use cases of the proposed loss: (1) networks that invert the Alex Net convolutional network; (2) a modified version of a variational autoencoder that generates realistic high-resolution random images.
Researcher Affiliation Academia Alexey Dosovitskiy and Thomas Brox University of Freiburg
Pseudocode No The paper describes model architectures and training procedures in text and tables but does not include any pseudocode or algorithm blocks.
Open Source Code No The paper does not provide a direct link to open-source code or an explicit statement that the code is publicly available.
Open Datasets Yes We trained on 227 227 pixel crops of images from the ILSVRC-2012 training set
Dataset Splits Yes We trained on 227 227 pixel crops of images from the ILSVRC-2012 training set and evaluated on the ILSVRC-2012 validation set.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU or CPU models, memory specifications) used for running the experiments.
Software Dependencies No We modified the caffe [23] framework to train the networks. For optimization we used Adam [25]... (Mentions software and optimizer names, but no version numbers for Caffe or specific libraries).
Experiment Setup Yes Coefficients for adversarial and image loss were respectively λadv = 100, λimg = 2 10 6. The feature loss coefficient λfeat depended on the comparator being used. It was set to 0.01 for the Alex Net CONV5 comparator... For optimization we used Adam [25] with momentum β1 = 0.9, β2 = 0.999 and initial learning rate 0.0002... We used batch size 64 in all experiments. The networks were trained for 500, 000-1, 000, 000 mini-batch iterations. In all networks we use leaky Re LU nonlinearities, that is, LRe LU(x) = max(x, 0) + α min(x, 0). We used α = 0.3 in our experiments.