nonlocalpooling 2.0.1

Global-Wise Pooling Layers for PyTorch

Non-Local Pooling

Update 2.0.1

• Now NonLocalPool2d will not use PixelShuffle by default since it may harm the performance. NonLocalPool3d still uses PixelShuffle3d to reduce the token number.

Update 1.2.0

• Now you can determine the output channel just like all the other learnable pooling methods (former version force out_channel=in_channel). However, if you specify out_channel, MaxPool would not work then since they cannot be added together. Leave out_channel to None to make the module works like before.

---

To use NonLocalPooling for your PyTorch project:

Step 1

pip install nonlocalpooling

Step 2

from nonlocalpooling.pool import NonLocalPool2d, NonLocalPool3d

---

Non-Local Pooling can be used to substitue your original PyTorch pooling methods.