You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+4-3Lines changed: 4 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -8,9 +8,10 @@ This is an implementation in PyTorch and this notebook assumes that you are alre
8
8
### What are Capsules?
9
9
10
10
Capsules are a small group of neurons that have a few key traits:
11
-
* Each neuron in a capsule represents various properties of a particular image part; properties like a parts color, width, etc.
12
-
* Every capsule **outputs a vector**, which has some magnitude and orientation.
13
-
* Capsules have a hierarchy between child and parent capsules and use **dynamic routing** to find the strongest connections between the output of one capsule and the inputs of the next layer of capsules.
11
+
* Each neuron in a capsule represents various properties of a particular image part; properties like a parts color, width, etc.
12
+
* Every capsule **outputs a vector**, which has some magnitude (that represents a part's **existence**) and orientation (that represents a part's generalized pose).
13
+
* A capsule network is made of multiple layers of capsules; during training, this network aims to learn the spatial relationships between the parts and whole of an object (ex. how the position of eyes and a nose relate to the position of a whole face in an image).
14
+
* Capsules represent relationships between parts of a whole object by using **dynamic routing** to weight the connections between one layer of capsules and the next and creating strong connections between spatially-related object parts.
0 commit comments