- Dense prediction and image classification are structurally different.
- In the work, we develop a new convolutional network module that is specifically designed for dense prediction.
- We show that the presented context module increases the accuracy of state-of-the-art semantic segmentation systems.
Read the full article, click here.
@karpathy: “Dilated convolutions are a very good idea. Expecting this to become standard already supported by Torch/Tensorflow”
[1511.07122] Multi-Scale Context Aggregation by Dilated Convolutions