Support Conv3D

3D convolutions are important too!

Like 2D images in nature, 3D Volume data is very common, such as medical CT and MRI data, where voxels are uniformly distributed in 3D space.

The medical and health field relies heavily on computer vision methods that have migrated from the field of 2D natural image processing, such as classification, segmentation and detection of lesions and organs. These methods also rely on convolutional neural networks, even in the Transformer of the VIT family, 3D convolution is an elegant and efficient implementation of 3DPatchEmbed.

Medical health is also a promising research direction of artificial intelligence, and has a bright application prospect. So, the 3D version of convolution should not be absent from our project.