EVOLVE: Efficient Learned Volume Compression with Variable-Rate Encoding on a Cross-Domain Database
EVOLVE uses one trained autoencoder to vary compression rates at inference time for scientific volume data.
The paper says the system was trained around a cross-domain database of 6,376 volumes from 21 simulations, curated for diversity with perceptual hashing. Its authors position EVOLVE against conventional compressors that lose fine structure at high compression ratios and INR methods that need expensive per-volume optimization. In tests on unseen simulation datasets, they report higher compression ratios at similar reconstruction quality and compression speeds far faster than INR-based methods. Code, model weights, and results are listed as available on the project page. ArXiv · AI/CL/LG's note
The paper says the system was trained around a cross-domain database of 6,376 volumes from 21 simulations, curated for diversity with perceptual hashing. Its authors position EVOLVE against conventional compressors that lose fine structure at high compression ratios and INR methods that need expensive per-volume optimization. In tests on unseen simulation datasets, they report higher compression ratios at similar reconstruction quality and compression speeds far faster than INR-based methods. Code, model weights, and results are listed as available on the project page. ArXiv · AI/CL/LG's note
score 4