Stefano Esposito
@s-esposito.bsky.social
🎮 Try it yourself!
Our WebGL viewer loads the baked meshes & textures right in the browser—runs on iPhone, Android, and Intel UHD laptops.
🎮 autonomousvision.github.io/volsurfs/vie...
Our WebGL viewer loads the baked meshes & textures right in the browser—runs on iPhone, Android, and Intel UHD laptops.
🎮 autonomousvision.github.io/volsurfs/vie...
May 5, 2025 at 1:00 PM
🎮 Try it yourself!
Our WebGL viewer loads the baked meshes & textures right in the browser—runs on iPhone, Android, and Intel UHD laptops.
🎮 autonomousvision.github.io/volsurfs/vie...
Our WebGL viewer loads the baked meshes & textures right in the browser—runs on iPhone, Android, and Intel UHD laptops.
🎮 autonomousvision.github.io/volsurfs/vie...
📱 Evaluation
Device : Samsung A52s (2021) − 720 p
• 65 FPS (3‑Mesh) → 35 FPS (9‑Mesh)
• Beats 3DGS (~ 8 FPS) and MobileNeRF (~ 24 FPS)
Image quality: 34.5 dB PSNR (7‑Mesh) — higher than MobileNeRF (29.3 dB) and close to 3DGS (35.4 dB).
Device : Samsung A52s (2021) − 720 p
• 65 FPS (3‑Mesh) → 35 FPS (9‑Mesh)
• Beats 3DGS (~ 8 FPS) and MobileNeRF (~ 24 FPS)
Image quality: 34.5 dB PSNR (7‑Mesh) — higher than MobileNeRF (29.3 dB) and close to 3DGS (35.4 dB).
May 5, 2025 at 1:00 PM
📱 Evaluation
Device : Samsung A52s (2021) − 720 p
• 65 FPS (3‑Mesh) → 35 FPS (9‑Mesh)
• Beats 3DGS (~ 8 FPS) and MobileNeRF (~ 24 FPS)
Image quality: 34.5 dB PSNR (7‑Mesh) — higher than MobileNeRF (29.3 dB) and close to 3DGS (35.4 dB).
Device : Samsung A52s (2021) − 720 p
• 65 FPS (3‑Mesh) → 35 FPS (9‑Mesh)
• Beats 3DGS (~ 8 FPS) and MobileNeRF (~ 24 FPS)
Image quality: 34.5 dB PSNR (7‑Mesh) — higher than MobileNeRF (29.3 dB) and close to 3DGS (35.4 dB).
🔧 How it works
We learn a fixed number of SDF “shells”.
1️⃣ Reconstruct implicit semi‑transparent adaptively-spaced shells.
2️⃣ Bake each shell into a tiny mesh + SH texture.
The GPU then just rasterizes and alpha‑blends in a fixed order—no ray‑marching, no depth sort.
We learn a fixed number of SDF “shells”.
1️⃣ Reconstruct implicit semi‑transparent adaptively-spaced shells.
2️⃣ Bake each shell into a tiny mesh + SH texture.
The GPU then just rasterizes and alpha‑blends in a fixed order—no ray‑marching, no depth sort.
May 5, 2025 at 1:00 PM
🔧 How it works
We learn a fixed number of SDF “shells”.
1️⃣ Reconstruct implicit semi‑transparent adaptively-spaced shells.
2️⃣ Bake each shell into a tiny mesh + SH texture.
The GPU then just rasterizes and alpha‑blends in a fixed order—no ray‑marching, no depth sort.
We learn a fixed number of SDF “shells”.
1️⃣ Reconstruct implicit semi‑transparent adaptively-spaced shells.
2️⃣ Bake each shell into a tiny mesh + SH texture.
The GPU then just rasterizes and alpha‑blends in a fixed order—no ray‑marching, no depth sort.
🚀 Efficiency and quality
• Surface rendering → fast but miss fuzz.
• Volume rendering → high capacity but not efficient.
• We skip sorting, skip empty space and samples number (3‑9) that are found by rasterizing smooth, lightweight meshes → quality approaching 3DGS, much higher rendering efficiency.
• Surface rendering → fast but miss fuzz.
• Volume rendering → high capacity but not efficient.
• We skip sorting, skip empty space and samples number (3‑9) that are found by rasterizing smooth, lightweight meshes → quality approaching 3DGS, much higher rendering efficiency.
May 5, 2025 at 1:00 PM
🚀 Efficiency and quality
• Surface rendering → fast but miss fuzz.
• Volume rendering → high capacity but not efficient.
• We skip sorting, skip empty space and samples number (3‑9) that are found by rasterizing smooth, lightweight meshes → quality approaching 3DGS, much higher rendering efficiency.
• Surface rendering → fast but miss fuzz.
• Volume rendering → high capacity but not efficient.
• We skip sorting, skip empty space and samples number (3‑9) that are found by rasterizing smooth, lightweight meshes → quality approaching 3DGS, much higher rendering efficiency.
📢 New paper CVPR 25!
Can meshes capture fuzzy geometry? Volumetric Surfaces uses adaptive textured shells to model hair, fur without the splatting / volume overhead. It’s fast, looks great, and runs in real time even on budget phones.
🔗 autonomousvision.github.io/volsurfs/
📄 arxiv.org/pdf/2409.02482
Can meshes capture fuzzy geometry? Volumetric Surfaces uses adaptive textured shells to model hair, fur without the splatting / volume overhead. It’s fast, looks great, and runs in real time even on budget phones.
🔗 autonomousvision.github.io/volsurfs/
📄 arxiv.org/pdf/2409.02482
May 5, 2025 at 1:00 PM
📢 New paper CVPR 25!
Can meshes capture fuzzy geometry? Volumetric Surfaces uses adaptive textured shells to model hair, fur without the splatting / volume overhead. It’s fast, looks great, and runs in real time even on budget phones.
🔗 autonomousvision.github.io/volsurfs/
📄 arxiv.org/pdf/2409.02482
Can meshes capture fuzzy geometry? Volumetric Surfaces uses adaptive textured shells to model hair, fur without the splatting / volume overhead. It’s fast, looks great, and runs in real time even on budget phones.
🔗 autonomousvision.github.io/volsurfs/
📄 arxiv.org/pdf/2409.02482