Switchable Atrous Convolution (SAC) softly switches the convolutional computation between different atrous rates and gathers the results using switch functions. The switch functions are spatially dependent, i.e., each location of the feature map might have different switches to control the outputs of SAC. To use SAC in a detector, we convert all the standard 3x3 convolutional layers in the bottom-up backbone to SAC.
Source: DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous ConvolutionPaper | Code | Results | Date | Stars |
---|
Task | Papers | Share |
---|---|---|
Reinforcement Learning (RL) | 76 | 38.97% |
Continuous Control | 24 | 12.31% |
Atari Games | 7 | 3.59% |
OpenAI Gym | 7 | 3.59% |
Autonomous Driving | 5 | 2.56% |
Decision Making | 5 | 2.56% |
Imitation Learning | 4 | 2.05% |
Meta-Learning | 3 | 1.54% |
Semantic Segmentation | 2 | 1.03% |
Component | Type |
|
---|---|---|
1x1 Convolution
|
Convolutions | |
Average Pooling
|
Pooling Operations | |
Convolution
|
Convolutions | |
Dilated Convolution
|
Convolutions | |
Global Average Pooling
|
Pooling Operations |