Lecture 27: Inverse Graphics/Research In Angjoo's Group (26)
SainanChen
Can it also be used for video/motion/object classification?
yang991178
Presumably yes, but feeding 3D objects to neural networks may not be as efficient or effective as inputing the original 2D images. 3D convolutions can be costly and we're essentially forcing the structure of the neural network into 2 separate stages. However, adding a discriminator network for classification could potentially improve the reconstruction network.
Can it also be used for video/motion/object classification?
Presumably yes, but feeding 3D objects to neural networks may not be as efficient or effective as inputing the original 2D images. 3D convolutions can be costly and we're essentially forcing the structure of the neural network into 2 separate stages. However, adding a discriminator network for classification could potentially improve the reconstruction network.