Generation
Asherchi opened this issue · comments
Hi, thanks for your work, which is definitly amazing. I want know whether the model can apply to another scene without fine-tune?
Hi @Asherchi, thank you for appreciating our work. For Neural Point-Based Graphics, finetune of the model itself is not absolutely necessary but optimizing descriptors for a new scene is a must -- otherwise it's not capable of rendering the new scene. Other works (espeically the NeRF track and others such as Instant Neural Graphics Primitives) introduce significant improvements in terms of the adaptation speed for a new scene.
An update here: a new NPBG++ can predict the descriptors for a new scene in a feedforward way.