diff options
author | 2016-09-29 15:05:32 -0800 | |
---|---|---|
committer | 2016-09-29 16:17:09 -0700 | |
commit | 1283b84a49a9f5e14aca833cf981b61848aaf916 (patch) | |
tree | ffba9d2d8ba549bd5981cc84748d2db8858fc676 /tensorflow/examples/tutorials/deepdream | |
parent | ef9f5fee0a079f6bed445064e8e9d18fb7a904d8 (diff) |
Merge changes from github.
Change: 134721831
Diffstat (limited to 'tensorflow/examples/tutorials/deepdream')
-rw-r--r-- | tensorflow/examples/tutorials/deepdream/deepdream.ipynb | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/tensorflow/examples/tutorials/deepdream/deepdream.ipynb b/tensorflow/examples/tutorials/deepdream/deepdream.ipynb index bb6d70d5c6..cbcc54ce3c 100644 --- a/tensorflow/examples/tutorials/deepdream/deepdream.ipynb +++ b/tensorflow/examples/tutorials/deepdream/deepdream.ipynb @@ -623,7 +623,7 @@ "<a id=\"laplacian\"></a>\n", "## Laplacian Pyramid Gradient Normalization\n", "\n", - "This looks better, but the resulting images mostly contain high frequencies. Can we improve it? One way is to add a smoothness prior into the optimization objective. This will effectively blur the image a little every iteration, suppressing the higher frequencies, so that the lower frequencies can catch up. This will require more iterations to produce a nice image. Why don't we just boost lower frequencies of the gradient instead? One way to achieve this is through the [Laplacian pyramid](https://en.wikipedia.org/wiki/Pyramid_%28image_processing%29#Laplacian_pyramid) decomposition. We call the resulting technique _Laplacian Pyramid Gradient Normailzation_." + "This looks better, but the resulting images mostly contain high frequencies. Can we improve it? One way is to add a smoothness prior into the optimization objective. This will effectively blur the image a little every iteration, suppressing the higher frequencies, so that the lower frequencies can catch up. This will require more iterations to produce a nice image. Why don't we just boost lower frequencies of the gradient instead? One way to achieve this is through the [Laplacian pyramid](https://en.wikipedia.org/wiki/Pyramid_%28image_processing%29#Laplacian_pyramid) decomposition. We call the resulting technique _Laplacian Pyramid Gradient Normalization_." ] }, { |