2024.12.04 - #18 - L3DG, IG-SLAM, ULSR-GS, ROVER, World Labs demo, MVD2, FastSR-NeRF, Zero-to-Hero

[changh95]

Interesting papers

𝐋𝟑𝐃𝐆: 𝐋𝐚𝐭𝐞𝐧𝐭 𝟑𝐃 𝐆𝐚𝐮𝐬𝐬𝐢𝐚𝐧 𝐃𝐢𝐟𝐟𝐮𝐬𝐢𝐨𝐧

• https://barbararoessle.github.io/l3dg/

IG-SLAM: Instant Gaussian SLAM

• https://arxiv.org/pdf/2408.01126](https://t.co/lj61ZN6RsV
• Dense RGB-only SLAM, 10 FPS
• DROID-SLAM backend (just like Droid-Splat like week!)
• Code: https://github.com/Liouvi/IG-SLAM

ULSR-GS: Ultra Large-scale Surface Reconstruction Gaussian Splatting with Multi-View Geometric Consistency

• https://ulsrgs.github.io

ROVER: A Multi-Season Dataset for Visual SLAM

• https://arxiv.org/pdf/2412.02506
• 실외 / 잔디깎기 로봇들을 위한 multi-weather 데이터셋

---

Industry

World Labs demo

• Fei-Fei Li 교수님의 스타트업
• https://youtu.be/9schOFFZtjs?si=pe54HqRnUZ1Y2PYP

---

박사 지원 팁

CS PhD SOP

• https://cs-sop.notion.site/CS-PhD-Statements-of-Purpose-df39955313834889b7ac5411c37b958d
• 미국/유럽 CS 박사과정 지원 시 SOP 예시 모음

[jy6757]

View Synthesis Optimization

MVD2 : Efficient Multiview 3D Reconstruction for Multiview Diffusion (Microsoft)
https://arxiv.org/abs/2402.14253

---

FastSR-NeRF: Improving NeRF Efficiency on Consumer Devices with A Simple Super-Resolution Pipeline (Apple)
https://openaccess.thecvf.com/content/WACV2024/papers/Lin_FastSR-NeRF_Improving_NeRF_Efficiency_on_Consumer_Devices_With_a_Simple_WACV_2024_paper.pdf

---

Zero-to-Hero: Enhancing Zero-Shot Novel View Synthesis via Attention Map Filtering (NVIDIA)
https://arxiv.org/abs/2405.18677