Deep Multi-Layer Perceptrons for Neural Reinforcement Learning

Home

Discover

Bookmarks

Keep Up With Latest Trending Papers. Computer Science, AI and Machine Learning and more.Subscribe

Liked byChristian and Jiayuan

CC BY-SA

Source arXiv

Computer Vision

Deep Multi-Layer Perceptrons for Neural Reinforcement Learning

KiloNeRF: Speeding up Neural Radiance Fields with Thousands of Tiny MLPs

Neural radiance view (nerf) synthesizes novel views of a scene with unprecedented quality by fitting a neural radiance field to rgb images.

However, the original model requires querying a deep multi-layer perceptron (mlp) millions of times, leading to slow renderingtimes even on modern graphics cards.

In our setting, each individual individual mlp only needs to represent parts of the scene, thus smaller and faster-to-evaluate mlps can be used.

By combining this divide-and-conquer strategy with further optimizations, rendering is accelerated by two orders of magnitude compared to the original model without incurring high storage costs.

Furthermore, using teacher-student distillation for training, we show that this speed-up can be achieved without sacrificing visual quality.

Neural radiance view (nerf) is a novel method for synthesizing novel views of a scene with unprecedented quality by fitting a neural radiance field to rgb images.

However, the original model requires querying a deep multi-layer perceptron (mlp) millions of times, leading to slow renderingtimes even on modern graphics cards.

Smaller and faster-to-evaluate mlps can be used.

Authors

Christian Reiser, Songyou Peng, Yiyi Liao, Andreas Geiger

◐ Recommended Follows

◐ Latest Activity