Merge pull request FluxML#64 from sambitdash/master

Changes to CNN to improve model accuracy.

vision/mnist/conv.jl

@@ -8,9 +8,9 @@ imgs = MNIST.images()
 
 labels = onehotbatch(MNIST.labels(), 0:9)
 
-# Partition into batches of size 1,000
+# Partition into batches of size 32
 train = [(cat(float.(imgs[i])..., dims = 4), labels[:,i])
-         for i in partition(1:60_000, 1000)]
+         for i in partition(1:60_000, 32)]
 
 train = gpu.(train)
 
@@ -19,12 +19,14 @@ tX = cat(float.(MNIST.images(:test)[1:1000])..., dims = 4) |> gpu
 tY = onehotbatch(MNIST.labels(:test)[1:1000], 0:9) |> gpu
 
 m = Chain(
-  Conv((2,2), 1=>16, relu),
-  x -> maxpool(x, (2,2)),
-  Conv((2,2), 16=>8, relu),
-  x -> maxpool(x, (2,2)),
-  x -> reshape(x, :, size(x, 4)),
-  Dense(288, 10), softmax) |> gpu
+    Conv((3, 3), 1=>32, relu),
+    Conv((3, 3), 32=>32, relu),
+    x -> maxpool(x, (2,2)),
+    Conv((3, 3), 32=>16, relu),
+    x -> maxpool(x, (2,2)),
+    Conv((3, 3), 16=>10, relu),
+    x -> reshape(x, :, size(x, 4)),
+    Dense(90, 10), softmax) |> gpu
 
 m(train[1][1])