tania
Reputation: 67
About downsampling in Unet segmentation architecture
Downsampling helps to reduce the shape but why downsampling is needed in Unet segmentation architecture?
Upvotes: 0
Views: 2059
Answers (1)
JoOkuma
Reputation: 492
Most of modern convolution neural network architectures use max-pooling to downsample (such as U-Net), not depending only on the stride to achieve this because pooling introduces a small location invariance to the network architecture and is faster to compute.
This might provide you more information.
Upvotes: 3
Related Questions
- How do we resize the segmented mask to Any size using UNET Neural Network Architecture
- Unet Segmentation Model
- Binary Cross Entropy Loss for Image Segmentation
- Concatenation step of U-Net for unequal number of channels
- How relevant are negative examples for a Unet segmentation model?
- Constructing Unet with pretrained Resnet34 encoder with Keras
- Deep Learning UNet convergence
- Why Reshape and Permute for segmentation with unet?
- Zero Jaccard accuracy in u-net implementation with keras
- Upsampling in Semantic Segmentation