Skip to content
This repository has been archived by the owner on Jan 2, 2021. It is now read-only.

Residual #129

Open
wants to merge 60 commits into
base: master
Choose a base branch
from
Open

Residual #129

wants to merge 60 commits into from

Conversation

ARDAKILIC2321
Copy link

lets

alexjc and others added 30 commits April 28, 2016 20:42
@alexjc
Experiments with new convolutional architectures.
fb0db99
@alexjc
Integrated new model, works but doesn't have enough capacity?
270524d
@alexjc
Merge commit '23a99313f975a8c0a41e8d05ee448cc4e8c50411' into forward
a4713ab 
Conflicts:
	doodle.py
@alexjc
Prototype of feed-forward architecture.
1bcbf2c
@alexjc
Re-introducing support for multiple phases, multiple iterations.
ffb7211
@alexjc
Better default options. 2 phases, 2 iterations.
54070f9
@alexjc
Switching architectures and newly trained network, preparing integrat…
94004e2 
…ion of weighting.
@alexjc
Prototype incrementally propagating data down decoder to the output.
abaab87
@alexjc
Layerwise decoding of the encoded + matched features, work in progress.
bead8a3
@alexjc
New balance parameter replaces both weights. Balance at 0.0 means ful…
d77ed33 
…l content, 1.0 means full style.
@longears @alexjc
Fix crash when a directory named "content" exists (#88)
7031555
@alexjc
Cleaning up multi-parameter implementation, supporting patch shapes.
419d752
@alexjc
Snap to grid depends on layers. Closes #87.
9af43f6
@alexjc
Reworking arguments to support nargs and lists, adding model configur…
9c7125a 
…ation for 5_1 to load new data.
@alexjc
Support for network that goes all the way up to 6_1.
5fbc79f
@alexjc
Layer-wise generation with micro-iterations to match patches.
f3edeab
@alexjc
Fully switching to a layerwise rather than iterative approach, removi…
9a188b8 
…ng un-used code.
@alexjc
Simplifications to the iterations, better debug logging, reduce load …
72361f8 
…times.
@alexjc
Refined the logging, removed unused seeding code.
9af96f5
@alexjc
Major improvements to display, ASCII logo too! All code in forward br…
fbb318f 
…anch is now AGPL.
@alexjc
Merge commit 'b2fac92b27dea18cf1ab8438c981b3537fa9ad06' into forward
3354fdb
@alexjc
Message to download the new model file as uncompressed.
7c5374f
@alexjc
Re-implementing visualization of intermediate frames, 50% slower.
35daa9a
@alexjc
The blending occurs every frame rather than between phases, fixing th…
da1c3d6 
…e --balance parameter.
@alexjc
Tidying up the code, comments, docker build.
fef8036
@alexjc
Re-ordered code, removed loss calculation, improved logging.
57a5e24
@alexjc
Updating Lasagne to it includes Deconv2D, adding sklearn to requireme…
a9377f9 
…nts.
@alexjc
Fix for image resize by using PIL fit() rather than thumbnail.
c0f9c69
@alexjc
Updating examples in the README after testing.
5b7c28f
@alexjc
New photo example.
e07a699
alexjc added 30 commits May 16, 2016 22:19
@alexjc
Replacing the balance parameter with content-weight and new paramet…
c784dd9 
…ers based on suggestion called `previous-weight` and `noise-weight`, each can be specified as arrays on the command-line. The blending is done after matching the patches.
@alexjc
Minor tweaks to filenames, code layout, adding prototype content reno…
6358992 
…rmalization but commented out.
@alexjc
Reworking layerwise code to allow for inter-layer iterations.
0d8cf11
@alexjc
Separate processing of layers.
833889f
@alexjc
Averaging the features from other layers before doing the next iterat…
feebb85 
…ion.
@alexjc
Re-introducing logging and command-line arguments.
8eddd55
@alexjc
Support for semantic maps again.
bc10b01
@alexjc
Support for arbitrary layer numbers in feature exchange code.
c1ddad5
@alexjc
Re-ordering of the exchange/merge operations.
ac03bad
@alexjc
Switch to new network architecture, simplified layer handling with in…
1f969b7 
…tegers.
@alexjc
Support for micro-iterations within the macro-passes.
f81446a
@alexjc
Integrated the feature merging code with the layer evaluation, can us…
90f7db3 
…e multiple iteration within layers again--helps with speed.
@alexjc
Cleaned up default parameters for testing.
288f3ac
@alexjc
First prototype of patch-matching.
ec1d401
@alexjc
Improving patch-matching performance.
1c08f62
@alexjc
Slow version of 3x3 patch-matching.
c4e6ceb
@alexjc
Minor performance improvements, major memory improvements. Sub-tensor…
99ec0a7 
… indexing seems to use a Python function on CPU, possibly offloaded to GPU with CUDA though.
@alexjc
Increased iteration count for PatchMatch, new network size with poten…
f07a188 
…tially better training.
@alexjc
Cleaning up patch-matching, stops at 10% improve threshold.
3a0418e
@alexjc
Working layerwise generation with exchanges between layers as soon as…
21da45f 
… possible, making sure information propagates down immediately as in feed-forward model.
@alexjc
Switch to JIT-compiled patch-matching.
8ca4787
@alexjc
Improved patch-matching using numba and JIT-compiled gu-functions.
3f81d50
@alexjc
Using previous layer's matched patches, restored display statistics.
ebe133e
@alexjc
Normalization of content features and using previous pass for patch m…
4015ade 
…atching.
@alexjc
Extracted patch score calculation, removed unused code.
b413689
@alexjc
Simplified model code, preparing for custom patch biases.
3eb2fe5
@alexjc
Experimental visualizations of the network.
ac57d04
@alexjc
Patch-variety experimental code using statistics, works in a single p…
6988155 
…ass but benefits from iteration still.
@alexjc
Removed content feature normalisation, prototype for matching style g…
f08c4d2 
…rams not just diversity, improved patch-matching, saving fewer files in debug mode.
@alexjc
Experiment with residual network. Features are not as well suited to …
fae15a7 
…expressing style.
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@ARDAKILIC2321 @alexjc @longears