Max-Heinrich Laves
Max-Heinrich Laves
I'm trying to scale the rigged hand using following code: ``` (window.controller = new Leap.Controller) .use('riggedHand', { scale: 5.0 }) .connect() ``` Hand shows and works correctly, but scaling doesn't...
Thanks for the nice framework! However, the medical imaging community is missing 3d operations, such as `conv3d`.
I would like to implement InstanceNorm3D. Given that BatchNorm3D and InstanceNorm2D are available as MIL op, this should be straightforward. However, it's not clear to me where I would start...
Conv3d
## Proposed changes * Implementation of naive `slow_conv_3D` on the cpu. * Implementation of `explicit_gemm_conv_ND_cpu`. * However, using this seems to be considerably slower than the naive implementation. I guess...
RDM does not compile for current versions of macOS. I created a pull request to fix that: https://github.com/avibrazil/RDM/pull/51
RDM now compiles natively for arm64 (Apple M1, Apple Silicon) Fixes https://github.com/avibrazil/RDM/issues/55
## Proposed changes This implementation extends `conv_general` and introduces a bool flag `transpose` to make use of the existing code path. * Includes tests and benchmarks. * Only supports groups=1...
## Proposed changes Added group support to `conv3d` forward pass (backward pass still missing, as for `conv2d`). I adapted `slow_conv_3D`, `explicit_gemm_conv_ND_cpu`, and `conv_3D_gpu` to support `groups > 1` and added...