Multi-Band Galaxy Novel-View Synthesis with Radiative Gaussian Fields
Organizations: University College London, UK · Baidu · AI Centre, University College London, UK
Abstract
We propose a novel view synthesis tool for galaxies based on Gaussian Splatting (GS). Unlike RGB data, where GS has achieved success, galaxy observations consist of multi-band signals from a massive number of bodies, contain complex patterns of cross-band interactions, and capture light propagation in highly heterogeneous media. As a result, the underlying physical process poses great challenges in model representation and optimization, which have not been investigated in existing research. To this end, we propose a physically inspired Gaussian Splatting method, GalaxyGS, which is equipped with physically meaningful components that capture both the energy emission mechanism of different astronomical bodies and the light propagation between them. In the absence of directly comparable baseline methods, we adopt a wide range of possible alternative approaches. Comparisons conclusively demonstrate that GalaxyGS is better in accuracy and physical fidelity across datasets and metrics.
Figures & tables
| Method | NMSE | ||||||
| FeatSplat-MB [ 27 ] | 0.516±0.154 | 0.328±0.041 | 0.336±0.126 | 0.186±0.089 | 0.121±0.051 | 6.817±3.613 | 0.107±0.049 |
| 3DGS-MB [ 18 ] | 0.354±0.114 | 0.261±0.057 | 0.256±0.072 | 0.145±0.044 | 0.094±0.044 | 4.789±2.396 | 0.093±0.035 |
| LGDWT-GS-MB [ 33 ] | 0.398±0.146 | 0.258±0.058 | 0.255±0.084 | 0.142±0.042 | 0.095±0.048 | 4.686±2.340 | 0.095±0.044 |
| MS-Splatting [ 28 ] | 2.684±5.069 | 0.630±0.126 | 0.397±0.106 | 0.342±0.141 | 0.300±0.144 | 12.216±2.153 | 0.181±0.118 |
| TensoRF-VM-MB [ 4 ] | 0.838±0.492 | 0.478±0.100 | 0.367±0.195 | 0.140±0.059 | 0.202±0.144 | 10.218±4.714 | 0.163±0.097 |
| 3DSatellites-MB [ 3 ] | 0.158±0.059 | 0.451±0.079 | 0.321±0.124 | 0.162±0.049 | 0.082±0.030 | 8.151±3.444 | 0.172±0.079 |
| Variant | NMSE | |||
| Shared | 0.203 | 0.265 | 0.140 | 2.451 |
| Direct stellar | 0.167 | 0.166 | 0.197 | 2.217 |
| Direct dust | 0.235 | 0.141 | 0.115 | 1.332 |
| Separate MIR/FIR | 0.060 | 0.035 | 0.034 | 0.494 |
| No attenuation field | 0.069 | 0.036 | 0.035 | 0.566 |
| One across bands | 0.069 | 0.035 | 0.035 | 0.564 |
Novel View Synthesis
Novel View Synthesis creates images of a scene from camera viewpoints not present in the input, using learned geometry and appearance models.
See all 317 papers