Added TimeDistributed option

raghakot · Nov 21, 2017 · a3bff40 · a3bff40
1 parent 5e9bcca
commit a3bff40
Show file tree

Hide file tree

Showing 2 changed files with 246 additions and 78 deletions.
diff --git a/README.md b/README.md
@@ -39,4 +39,16 @@ Includes cifar10 training example. Achieves ~86% accuracy using Resnet18 model.
 
 Note that ResNet18 as implemented doesn't really seem appropriate for CIFAR-10 as the last two residual stages end up 
 as all 1x1 convolutions from downsampling (stride). This is worse for deeper versions. A smaller, modified ResNet-like 
-architecture achieves ~92% accuracy (see [gist](https://gist.github.com/JefferyRPrice/c1ecc3d67068c8d9b3120475baba1d7e)). 
+architecture achieves ~92% accuracy (see [gist](https://gist.github.com/JefferyRPrice/c1ecc3d67068c8d9b3120475baba1d7e)).
+
+## Update: 2017-11-21
+
+### TimeDistributed
+
+Many CNN+LSTM architectures require a TimeDistributed version of CNNs, which are applied on time-distributed inputs as common filter weights, and then fed to an LSTM. Invoking the argument
+
+```py
+time_distributed=True 
+```
+
+creates a TimeDistributed version of resnet. The input_shape must be in the form `(nb_time_steps, nb_rows, nb_cols, nb_channels)`.