RDGSplat: Render-Dedicated Geometry for Novel View Synthesis
To this end, we propose RDGSplat, a framework that decodes a second geometry dedicated to rendering from a frozen 3D foundation model, leaving its metric predictions intact.
ProofPaper ↗
Key points
- 3D foundation models enable efficient novel view synthesis by carrying a Gaussian head on the representation they already use for reconstruction.
- In particular, we devise Render-Dedicated Geometry Decoding, which duplicates the pretrained decoders and optimizes the duplicates under photometric supervision alone.
- Then, a Target-Pose Conditioned Adapter is introduced to reformulate the representation those decoders read, conditioned on the target camera pose rather than the target image.
- Extensive experiments show that RDGSplat improves novel view synthesis across three feed-forward backbones on four benchmarks, with every pretrained weight frozen.
Sources (1)
- [1]RDGSplat: Render-Dedicated Geometry for Novel View SynthesisarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 10:18 PM
To this end, we propose RDGSplat, a framework that decodes a second geometry dedicated to rendering from a frozen 3D foundation model, leaving its metric predictions intact.
3D foundation models enable efficient novel view synthesis by carrying a Gaussian head on the representation they already use for reconstruction.
Extractive summary: sentences quoted from the sources.