MixRT: Mixed Neural Representations For Real-Time NeRF Rendering

Georgia Tech

Meta



Abstract Neural Radiance Field (NeRF) has emerged as a leading technique for novel view synthesis, owing to its impressive photorealistic reconstruction and rendering capability. Nevertheless, achieving real-time NeRF rendering in large-scale scenes has presented challenges, often leading to the adoption of either intricate baked mesh representations with a substantial number of triangles or resource-intensive ray marching in baked representations. We challenge these conventions, observing that high-quality geometry, represented by meshes with substantial triangles, is not necessary for achieving photorealistic rendering quality. Consequently, we propose MixRT, a novel NeRF representation that includes a low-quality mesh, a view-dependent displacement map, and a compressed NeRF model. This design effectively harnesses the capabilities of existing graphics hardware, thus enabling real-time NeRF rendering on edge devices. Leveraging a highly-optimized WebGL-based rendering framework, our proposed MixRT attains real-time rendering speeds on edge devices (over 30 FPS at a resolution of 1280 x 720 on a MacBook M1 Pro laptop), better rendering quality (0.2 PSNR higher in indoor scenes of the Unbounded-360 datasets), and a smaller storage size (less than 80% compared to state-of-the-art methods).


์‹ ๊ฒฝ ๋ณต์‚ฌ ํ•„๋“œ(NeRF)๋Š” ๋†€๋ผ์šด ์‚ฌ์‹ค์  ์žฌ๊ตฌ์„ฑ ๋ฐ ๋ Œ๋”๋ง ๋Šฅ๋ ฅ์œผ๋กœ ์ธํ•ด ์ƒˆ๋กœ์šด ์‹œ์  ํ•ฉ์„ฑ์„ ์œ„ํ•œ ์ฃผ์š” ๊ธฐ์ˆ ๋กœ ๋ถ€์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿผ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ๋Œ€๊ทœ๋ชจ ์žฅ๋ฉด์—์„œ ์‹ค์‹œ๊ฐ„ NeRF ๋ Œ๋”๋ง์„ ๋‹ฌ์„ฑํ•˜๋Š” ๊ฒƒ์€ ๋„์ „์ด ๋˜์—ˆ์œผ๋ฉฐ, ์ด๋Š” ์ข…์ข… ์ƒ๋‹นํ•œ ์ˆ˜์˜ ์‚ผ๊ฐํ˜•์„ ๊ฐ€์ง„ ๋ณต์žกํ•œ ๋ฒ ์ดํฌย ๋ฉ”์‹œ ํ‘œํ˜„ ๋˜๋Š” ๋ฒ ์ดํฌ๋œ ํ‘œํ˜„์—์„œ์˜ ์ž์› ์ง‘์•ฝ์ ์ธ ๋ ˆ์ด ๋งˆ์นญ์„ ์ฑ„ํƒํ•˜๋„๋ก ์ด๋Œ์—ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด๋Ÿฌํ•œ ๊ด€๋ก€์— ๋„์ „ํ•˜๋ฉฐ, ์ƒ๋‹นํ•œ ์‚ผ๊ฐํ˜•์„ ๊ฐ€์ง„ ๋ฉ”์‹œ๋กœ ํ‘œํ˜„๋œ ๊ณ ํ’ˆ์งˆ ๊ธฐํ•˜ํ•™์ด ์‚ฌ์‹ค์  ๋ Œ๋”๋ง ํ’ˆ์งˆ์„ ๋‹ฌ์„ฑํ•˜๊ธฐ ์œ„ํ•ด ํ•„์š”ํ•˜์ง€ ์•Š๋‹ค๋Š” ๊ฒƒ์„ ๊ด€์ฐฐํ–ˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ, ์ €ํ’ˆ์งˆ ๋ฉ”์‹œ, ์‹œ์  ์˜์กด์  ๋ณ€์œ„ ๋งต, ๊ทธ๋ฆฌ๊ณ  ์••์ถ•๋œ NeRF ๋ชจ๋ธ์„ ํฌํ•จํ•˜๋Š” ์ƒˆ๋กœ์šด NeRF ํ‘œํ˜„์ธ MixRT๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์ด ์„ค๊ณ„๋Š” ๊ธฐ์กด ๊ทธ๋ž˜ํ”ฝ ํ•˜๋“œ์›จ์–ด์˜ ๋Šฅ๋ ฅ์„ ํšจ๊ณผ์ ์œผ๋กœ ํ™œ์šฉํ•˜์—ฌ ์—์ง€ ์žฅ์น˜์—์„œ ์‹ค์‹œ๊ฐ„ NeRF ๋ Œ๋”๋ง์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ์ตœ์ ํ™”๋œ WebGL ๊ธฐ๋ฐ˜ ๋ Œ๋”๋ง ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ํ™œ์šฉํ•˜์—ฌ, ์šฐ๋ฆฌ์˜ ์ œ์•ˆํ•œ MixRT๋Š” ์—์ง€ ์žฅ์น˜์—์„œ ์‹ค์‹œ๊ฐ„ ๋ Œ๋”๋ง ์†๋„(๋งฅ๋ถ M1 Pro ๋…ธํŠธ๋ถ์—์„œ 1280 x 720 ํ•ด์ƒ๋„์—์„œ ์ดˆ๋‹น 30 ํ”„๋ ˆ์ž„ ์ด์ƒ)๋ฅผ ๋‹ฌ์„ฑํ•˜๊ณ , ๋” ๋‚˜์€ ๋ Œ๋”๋ง ํ’ˆ์งˆ(๋น„๊ต์  ํ–ฅ์ƒ๋œ PSNR)์„ ์ œ๊ณตํ•˜๋ฉฐ, ์ƒํƒœ ๊ธฐ์ˆ  ๋ฐฉ๋ฒ•์— ๋น„ํ•ด ์ €์žฅ ํฌ๊ธฐ๊ฐ€ ์ž‘์Šต๋‹ˆ๋‹ค(80% ๋ฏธ๋งŒ).







https://arxiv.org/abs/2312.11841