TorchMorph: CUDA-Accelerated Morphological Transforms for PyTorch
August 24, 2026
TorchMorph provides 22 fused CUDA kernels for morphological operations, including binary/grayscale morphology and distance transforms. This allows for direct GPU processing of (B, C, Spatial...) tensors, eliminating CPU-to-GPU bottlenecks found in scipy.ndimage.
HOW THIS AFFECTS YOU
●
builderYou can now integrate complex morphological transforms directly into your GPU training loops without expensive data transfers.