Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Ọrọ-si-3D Iran adanwo

Test DreamFusion distillation, NeRF optimization, Janus artifacts, feed-forward models, multi-view consistency, and 3D limits.

Jẹmọ itọsọna awọn ọnaỌrọ Lati 3D Iran
Ibeere 1 ti 8

Kini isọdọtun bọtini ti DreamFusion (2022)?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini isọdọtun bọtini ti DreamFusion (2022)?
    • O lo Iṣayẹwo Distillation Score lati ṣe 3D lati awoṣe itankale 2D laisi data ikẹkọ 3D
    • O jẹ akọkọ lati ṣe ọlọjẹ awọn ohun gidi pẹlu LiDAR
    • O ṣe ikẹkọ taara lori ipilẹ data mesh mesh 3D nla kan
    • O ṣe ipilẹṣẹ awọn awoṣe 3D lesekese ni lilo iwe-iwọle siwaju ẹyọkan
  2. Kini 'iṣoro Janus' ni ọrọ-si-3D?
    • Awọn awoṣe ṣe laiyara pupọ lati wulo
    • Awọn nkan ti ipilẹṣẹ fihan ọpọlọpọ awọn oju tabi awọn iwaju lati awọn igun oriṣiriṣi
    • Sojurigindin han fo jade ati grẹy
    • Ọna kika faili 3D ko ni ibamu pẹlu awọn ẹrọ ere
  3. Tọkọtaya wo ni OpenAI's ni kutukutu ọrọ-si-3D awọn awoṣe siwaju kikọ?
    • DALL-E ati CLIP
    • Itankale Iduroṣinṣin ati Midjourney
    • Point-E ati Shap-E
    • DreamFusion ati Magic3D
  4. Kini idi ti awọn ọna ti o da lori iṣapeye tete bii DreamFusion lọra?
    • Wọn ni lati ṣe igbasilẹ awọn awoṣe lati inu awọsanma ni akoko kọọkan
    • Wọn ṣe ni ipinnu 8K nipasẹ aiyipada
    • Wọn nilo fifi ọwọ ṣe laarin awọn igbesẹ
    • Wọn ṣe iṣapeye aṣoju 3D lọtọ fun ohun kan lori ọpọlọpọ awọn iterations Rendering
  5. Eyi ti o wu kika jẹ KO kan aṣoju 3D oniduro yi ni awọn ọna šiše?
    • MP3 iwe igbi fọọmu
    • Apapo polygon
    • Aaye Radiance (NeRF)
    • Ojuami awọsanma
  6. Kini 'iduro-iwoye-ọpọlọpọ' tọka si bi ipenija?
    • Ṣiṣe iwọn faili ni ibamu laarin awọn okeere
    • Aridaju pe ohun naa dabi ibaramu ati pe o tọ lati gbogbo igun kamẹra
    • Ntọju itọka ọrọ labẹ opin ọrọ
    • Ibamu paleti awọ si itọsọna ami iyasọtọ kan
  7. What representation does Shap-E directly generate?
    • Only a fixed-resolution RGB image
    • A text transcript with speaker labels
    • Parameters of implicit functions that can render as textured meshes or neural radiance fields
    • A single unstructured audio waveform
  8. How does the Point-E paper’s two-stage system produce a 3D point cloud?
    • A 3D artist manually models every prompt
    • A NeRF is optimized from scratch using hundreds of camera views
    • A classifier predicts a file extension without generating geometry
    • A text-to-image model first generates a synthetic view, then an image-conditioned diffusion model produces a point cloud