This repository was archived by the owner on Jan 2, 2021. It is now read-only.

Residual#129

Open

ARDAKILIC2321 wants to merge 60 commits intomasterfrom

Commits on Apr 28, 2016

Experiments with new convolutional architectures.
alexjc
committed

Commits on May 2, 2016

Integrated new model, works but doesn't have enough capacity?
alexjc
committed

Commits on May 4, 2016

Commits on May 7, 2016

Commits on May 10, 2016

Switching architectures and newly trained network, preparing integration of weighting.
alexjc
committed
Prototype incrementally propagating data down decoder to the output.
alexjc
committed

Commits on May 11, 2016

Commits on May 13, 2016

Commits on May 14, 2016

Commits on May 15, 2016

Commits on May 16, 2016

Commits on May 19, 2016

Minor tweaks to filenames, code layout, adding prototype content renormalization but commented out.
alexjc
committed

Commits on May 23, 2016

Commits on May 29, 2016

Commits on May 31, 2016

Commits on Jun 1, 2016

Improving patch-matching performance.
alexjc
committed

Commits on Jun 3, 2016

Commits on Jun 9, 2016

Commits on Jun 12, 2016

Working layerwise generation with exchanges between layers as soon as possible, making sure information propagates down immediately as in feed-forward model.
alexjc
committed

Commits on Jun 13, 2016

Switch to JIT-compiled patch-matching.
alexjc
committed

Commits on Jun 15, 2016

Simplified model code, preparing for custom patch biases.
alexjc
committed

Commits on Jun 16, 2016

Experimental visualizations of the network.
alexjc
committed

Commits on Jun 18, 2016

Patch-variety experimental code using statistics, works in a single pass but benefits from iteration still.
alexjc
committed

Commits on Jun 19, 2016

Removed content feature normalisation, prototype for matching style grams not just diversity, improved patch-matching, saving fewer files in debug mode.
alexjc
committed

Commits on Jul 3, 2016

Experiment with residual network. Features are not as well suited to expressing style.
alexjc
committed