Efficient Composing of Multiple Objects into Scene Representation

[kamzero]

Hi, I want to thank you for your excellent work. I'm impressed by the method you proposed for composing multiple objects into a scene representation.

Your approach of concatenating the features of each object along the rank dimension and warping the ray into each object's coordinate system before querying the density and color at the sampled points is effective. However, I have a question regarding the implementation details of this method. While this approach is effective, it also requires storing the representation of each object separately and concatenating them at rendering time, which can be computationally expensive and memory-intensive, especially for large scenes with many objects. I was wondering if there is any way to compose multiple transformed objects into a whole scene representation that is more memory-efficient and computationally efficient than concatenating the features of each object separately. Have you explored any other methods for composing multiple objects into a scene representation that could achieve this goal?

Thank you for your time and for sharing your research with the community. I'm looking forward to hearing your thoughts on this.

Best regards