Convolutional layers remain important for computer vision problems even after the transformer revolution, so they are important to have:
For these, we need to address pymc-devs/pytensor#548
In addition, convolutional network designs typically include some kind of pooling layers:
As well as helpers for padding images to the correct size:
A pt.pad function would be nice for implementing these; I opened an issue for that here pymc-devs/pytensor#743
Convolutional layers remain important for computer vision problems even after the transformer revolution, so they are important to have:
For these, we need to address pymc-devs/pytensor#548
In addition, convolutional network designs typically include some kind of pooling layers:
MaxUnpool2d— dropped: it serves SegNet-style decoders that transposed convolution has since replaced, torch is the only framework still shipping it, and it is already reachable aspt.grad(cost=None, wrt=X, known_grads={pooled: values}).As well as helpers for padding images to the correct size:
A
pt.padfunction would be nice for implementing these; I opened an issue for that here pymc-devs/pytensor#743