Megadose The AI curve, compressed.

Towards Seconds-level Video Representation: Quadratically Gated Feature Grid with Adaptive, Prediction-free Sampling

· ArXiv · AI/CL/LG ·
Q-GAS fits and quantizes three tested video sequences in under 50 seconds on a single RTX PRO 6000.

The authors report decoding above 370 frames per second with mixed precision and about 146 FPS in FP32 on UVG Beauty, Bosphorus and Jockey. The method combines explicit 3D feature grids with a compact quadratic-gating decoder, accepting modestly higher storage to improve expressive capacity. Its sampler mixes high-error locations with uniform samples without a separate prediction model. In their PyTorch implementation, training and quantization were 5.5–7.2 times faster than the evaluated Instant-NGP baseline and 19–28 times faster than RNeRV.

ArXiv · AI/CL/LG's note

score 4

Categories: Research