So far it’s been a productive autumn. Firstly, our framework, Cloning-Plenoxels (Clonoxels), was accepted as a short paper at the ACM SIGGRAPH European Conference on Visual Media Production 2025. Excited to present this approach to enhancing the efficiency of explicit radiance fields by automatically detecting and sharing information between repeat objects. As discussed in the paper, I really think that if this principle is pushed, we could see some interesting advances in novel view synthesis and object-based media. There are all kinds of ways in which repeated patterns or structures could be leveraged - making use of vision model feature embeddings would be cool!
In October, we submitted our paper, “Emergent Denoising of SDSS Galaxy Spectra Through Unsupervised Deep Learning”, to The Monthly Notices of the Royal Astronomical Society for review. This was a long time coming, making use of work I did way back during my Master’s research and completed over many weekends. We use unsupervised deep learning to improve spectral data quality while illustrating how classical denoising techniques can fail to recover underlying signals. We also show that related deep methods demonstrate promise in reconstructing spectra from incomplete data.
As for things to come, a paper is currently making its way out of the experimentation phase and new collaborations are being discussed. Hope that winter sees yet more science!