Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
ImplicitTerrainV2 is introduced as a compact and efficient neural representation for terrain data, addressing limitations of previous implicit neural representations (INRs) in GIS applications. It employs a wavelet complexity field (WCF) to derive spatially-adaptive frequency masks, focusing high-frequency capacity on complex terrain regions and guiding complexity-aware adaptive sampling. The model incorporates derivative-aware supervision and post-training compression, achieving a PSNR of 66.25 dB on Swiss terrain tiles with 3.2x fewer parameters and faster training compared to prior work, while also being competitive with established DEM codecs.
Terrain data can be represented with surprisingly high fidelity (66.25 dB PSNR) and compressed to 1.23 bits per pixel using a wavelet-guided neural implicit representation, outperforming previous methods in both accuracy and efficiency.
Digital elevation models (DEMs) underpin terrain analysis in Geographic Information Systems (GIS), but in their common raster form, they rely on interpolation for off-grid sampling and finite-difference operators for derivative-based analysis. Implicit neural representations (INRs) offer a continuous alternative, but prior terrain INRs lack explicit frequency control, neglect the gradient structure of terrain, and remain too large and costly to train for practical deployment. We present ImplicitTerrainV2, which advances terrain INRs toward a compact, efficient neural terrain data format by combining a spectral control mechanism with wavelet-guided spatial adaptivity, derivative-aware supervision, and post-training model compression. At its core, a wavelet complexity field (WCF) derives spatially-adaptive frequency masks from analytically computed wavelet coefficients, localizing high-frequency capacity to complex terrain regions. The same field guides complexity-aware adaptive sampling that concentrates training in high-complexity regions, while gradient matching applies extra supervision to enforce the smooth manifold structure of terrain DEMs for improved derivative fidelity. Post-training mixed-precision quantization and entropy coding reduce storage to 1.23 bpp with a 0.28 dB PSNR drop. On 50 Swiss terrain tiles, ImplicitTerrainV2 reaches 66.25 dB end-to-end PSNR, improving over the prior work by 5.70 dB while using 3.2x fewer parameters and training in 55 s per tile on a single GPU. Our compressed neural format is competitive with several established DEM codecs in rate-distortion performance, while additionally supporting off-grid point queries, closed-form derivative evaluation, and resolution-independent reconstruction, which may benefit many downstream GIS applications.