Перейти к содержимому

The Biggest Breakthrough in Gaussian Splatting - A 3DGRUT First Impressions and Tutorial

Pixel Reconstruct

0:00 / 0:00

The Biggest Breakthrough in Gaussian Splatting - A 3DGRUT First Impressions and Tutorial

33 073 просмотра · 1 г. назад
Pixel Reconstruct
20,5 тыс. подписчиков
33 073 просмотра · 1 г. назад
💥 Gaussian Splatting + Ray Tracing = 3DGRUT 🔥 Learn how NVIDIA’s new 3DGRUT project lets you create full 3D scenes using fisheye lenses and just 1/3 the number of images you'd need with traditional Gaussian Splatting. This first impressions and tutorial walks you through everything: setup, training, visualization, and exporting your own radiance fields. As well as showing you expected results with my own data. This is the most flexible and fisheye-friendly version of Gaussian Splatting yet — and it works with noisy images, fewer shots, and rolling shutter too. 👇 Jump to what you need 👇 ⏱️ Timestamps 00:00 Intro 00:35 3DGRUT Overview 04:21 Visualizing data and first impressions 11:58 The 3DGRUT Playground 16:29 Reviewing metrics and ply exports 09:10 Playground + Ray Tracing Effects 19:50 How to Install 3DGRUT 24:36 Preparing Your Image Data 30:06 Training the Model 34:00 Visualizing Results 36:53 00 Final Thoughts 🔗 Links & Resources 📌 My GitHub Fork with expanded instruction (Start here): https://github.com/jonstephens85/3dgrut 📌 Official NVIDIA 3DGRUT repo: https://github.com/nv-tlabs/3dgrut 📌 COLMAP download: https://colmap.github.io/ 📬 Subscribe for more real-world AI + 3D tutorials If you're into computer vision, 3D reconstruction, and cutting-edge tools like Gaussian Splatting and NeRFs — hit that Subscribe button and turn on notifications 🔔 so you never miss a drop. 👉    / @pixelreconstruct