1 minute read

Reference

Block-NeRF: Scalable Large Scene Neural View Synthesis

Large-scale scene reconstruction์—์„œ ์–ด๋ ค์šด ์ 

  • the presence of transient objects (cars and pedestrians)
  • highly unlikely to be collected in a single capture under consistent conditions
  • variance in scene geometry (e.g., construction work and parked cars)
  • variance in appearance (e.g., weather conditions and time of day)

Reconstructing such large-scale environments introduces additional challenges, including the presence of transient objects (cars and pedestrians), limitations in model capacity, along with memory and compute constraints.

Furthermore, training data for such large environments is highly unlikely to be collected in a single capture under consistent conditions.

Rather, data for different parts of the environment may need to be sourced from different data collection efforts, introducing variance in both scene geometry (e.g., construction work and parked cars), as well as appearance (e.g., weather conditions and time of day).

  • ๋Œ€๊ทœ๋ชจ ํ™˜๊ฒฝ ์žฌ๊ตฌ์„ฑ์—๋Š” ์—ฌ๋Ÿฌ ๋„์ „ ๊ณผ์ œ๊ฐ€ ์žˆ์Œ:
    • ์ผ์‹œ์  ๊ฐ์ฒด(์ž๋™์ฐจ, ๋ณดํ–‰์ž)์˜ ์กด์žฌ.
    • ๋ชจ๋ธ ์šฉ๋Ÿ‰, ๋ฉ”๋ชจ๋ฆฌ, ๊ณ„์‚ฐ ๋ฆฌ์†Œ์Šค์˜ ํ•œ๊ณ„.
  • ๋Œ€๊ทœ๋ชจ ํ™˜๊ฒฝ์˜ ํ•™์Šต ๋ฐ์ดํ„ฐ๋Š” ์ผ๊ด€๋œ ์กฐ๊ฑด์—์„œ ๋‹จ์ผ ์บก์ฒ˜๋กœ ์ˆ˜์ง‘๋˜๊ธฐ ์–ด๋ ต๊ณ ,
    • ๋‹ค๋ฅธ ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘ ๋…ธ๋ ฅ์—์„œ ๊ฐ๊ธฐ ๋‹ค๋ฅธ ํ™˜๊ฒฝ ๋ฐ์ดํ„ฐ๋ฅผ ์กฐํ•ฉํ•ด์•ผ ํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์Œ.
    • ์ด๋กœ ์ธํ•ด ์žฅ๋ฉด ๊ธฐํ•˜ํ•™(์˜ˆ: ๊ณต์‚ฌ, ์ฃผ์ฐจ๋œ ์ฐจ๋Ÿ‰)๊ณผ ์™ธํ˜•(์˜ˆ: ๋‚ ์”จ, ์‹œ๊ฐ„)์—์„œ ๋ณ€๋™์„ฑ ๋ฐœ์ƒ.

Block-NeRF์˜ ํ™•์žฅ ๋ฐ ํ•œ๊ณ„:

  • ๋„คํŠธ์›Œํฌ ์šฉ๋Ÿ‰์„ ํ™•์žฅํ•˜๋ฉด ๋” ํฐ ์žฅ๋ฉด์„ ํ‘œํ˜„ ๊ฐ€๋Šฅ.
    • ๊ทธ๋Ÿฌ๋‚˜ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์ œ์•ฝ ๋ฐœ์ƒ:
      • ๋ Œ๋”๋ง ์‹œ๊ฐ„์ด ๋„คํŠธ์›Œํฌ ํฌ๊ธฐ์— ๋น„๋ก€ํ•ด ์ฆ๊ฐ€.
      • ๋„คํŠธ์›Œํฌ๊ฐ€ ๋‹จ์ผ ์ปดํ“จํŒ… ์žฅ์น˜์— ์ ํ•ฉํ•˜์ง€ ์•Š์Œ.
      • ํ™˜๊ฒฝ ์—…๋ฐ์ดํŠธ๋‚˜ ํ™•์žฅ์€ ๋„คํŠธ์›Œํฌ ์ „์ฒด ์žฌํ•™์Šต(retraining)์ด ํ•„์š”.
  • ๋ฌธ์ œ ํ•ด๊ฒฐ ๋ฐฉ์•ˆ:
    • ๋Œ€๊ทœ๋ชจ ํ™˜๊ฒฝ์„ ๊ฐœ๋ณ„ Block-NeRF๋กœ ๋ถ„ํ•  ๋ฐ ๋…๋ฆฝ์  ํ•™์Šต.
    • ์ถ”๋ก  ์‹œ ๊ด€๋ จ๋œ Block-NeRF๋งŒ ๋ Œ๋”๋งํ•˜์—ฌ ๋™์ ์œผ๋กœ ๊ฒฐํ•ฉ ๋ฐ ์ƒ์„ฑ.
    • ์ด๋ฅผ ํ†ตํ•ด:
      • ์œ ์—ฐ์„ฑ ๊ทน๋Œ€ํ™”, ์ž„์˜์˜ ๋Œ€๊ทœ๋ชจ ํ™˜๊ฒฝ ์ฒ˜๋ฆฌ ๊ฐ€๋Šฅ.
      • ํ™˜๊ฒฝ์˜ ๋ถ€๋ถ„์  ์—…๋ฐ์ดํŠธ ๋ฐ ํ™•์žฅ ๊ฐ€๋Šฅ(์žฌํ•™์Šต ๋ถˆํ•„์š”).

Leave a comment