> What I find most exciting about it is that a NeRF represents images as neural nets, one neural net for each image (in the OP paper generalised to image + defo

> What I find most exciting about it is that a NeRF represents images as neural nets, one neural net for each image (in the OP paper generalised to image + deformations). By evaluating the net at various pixel coordinates it gives the color. I actually had a similar idea and used it to write a lossy image compressor for my bachelor thesis in 2018: https://theses.liacs.nl/pdf/2018-2019-PetersO.pdf https://theses.liacs.nl/pdf/2018-2019-PetersO.pdf