search

aharley / segaware

Segmentation-Aware Convolutional Networks Using Local Attention Masks
145 stars 31 forks source link

Is the TwoImageData layer missing in the repo? #2

Closed CHUNYUWANG closed 7 years ago

aharley commented 7 years ago

Ah! You can swap TwoImageData with ImageSegData. See the "res" example to see how to use ImageSegData. I'll fix this in the repo soon. Thanks for bringing it up. The issue came from developing the VGG-based stuff with DeepLabV1 code, and the ResNet-based stuff with DeepLabV2 code.

brisker commented 6 years ago

@aharley how did you visualize these embeddings since they are tensors in size like 64(or larger than 3) H W?

image image

aharley commented 6 years ago

PCA compression down to 3 dimensions.