This is a 3D model ("Mary") of an adult (curved carapace length 940 mm) female flatback sea turtle (Natator depressus, PIT tag 900119000520197, captured 11/21/17, Thevenard Island, Western Australia) as part of the Department of Biodiversity, Conservation and Attractions' North West Shelf Flatback Turtle Conservation Program (DBCA-NWSFTCP) with Scott Whiting (DBCA), Sabrina Fossette (DBCA) and Adrian Gleiss (Murdoch University). The turtle was photographed on a nesting beach using the Beastcam system and released in good condition. The 3D model was created through a photogrammetry scan using COLMAP and Autodesk REMAKE. Animator Jer Bot used Blender to reconstruct the sea turtle. Credit should be given to the Digital Life project and NWSFTCP: “The author(s) acknowledge the Digital Life project and the Department of Biodiversity, Conservation and Attractions’ North West Shelf Flatback Turtle Conservation Program for creation of the 3D flatback turtle model”.

Inquire about licensing: **www.digitallife3d.org**.

Model 53A - Flatback Sea Turtle

Licensed under **CC Attribution-NonCommercial-NoDerivs**

3D scan of the family house, in which my grandmother grew up!

It is located in the middle of fields in the "[Côte des Légendes](https://goo.gl/BhWdrr)" area in Brittany in France, and initially served as a farmhouse, a barn and a stable: there is still a massive granite trough inside the main building, which purpose was to feed the animals.

Processed from 670 pictures on Pocophone F1, some of them acquired with a 4.5m telescopic glass-washing handle, used as a (tall) DIY monopod. I still had to manually correct and paint some of the higher parts of the model for it to display nicely, as well as carve into the windows to improve on realism. The normals (Matcap + Surface) are not baked from bump maps, but actual geometry.

* Processed with Colmap, OpenMVS, Blender and BakeMyScan. 
* Background sound (which nicely depicts the location's atmosphere, some noisy pidgeons sometimes even nest under the roof!) from [Freesound.org](https://goo.gl/VPjh7v)

Country House

"Mokyo Night Walker"

Short 4k visual art animation using this point cloud is available for viewing/download here: https://www.youtube.com/watch?v=j_4D_oiJFQQ
Scan made with colmap using frames extracted from a video walk located at these coordinates:  https://youtu.be/uxTXXhBCl-Y?t=1646 
All geometry wrangling, animation and rendering performed with Houdini.  Special thanks to Ben at SideFX for directly inspiring this upload :)

Special thanks to Nippon Wandering TV for creating and publishing many wonderful walks for us to enjoy as time and space permit.  :)

Sea U Soon,
-Alan/Organic

This signal was analyzed by Organic.  
Additional data may be found here: 
https://www.instagram.com/p/CEjHJW9Hrwr/

Tools used: youtube-dl, ffmpeg, python, colmap, houdini, blender

https://artbyorganic.com

nippon_wandering_tokyo_nights_001_10m_pts

An "augmented" scan of the family's country house granite fireplace ([here](https://skfb.ly/6HSFy) is a scan of the building), which was the absolute "central piece" of the house when it was still a farm.

Although there is still room for improvement on the modeling and rendering, I've tried to "upgrade" the raw scan in Blender by adding some firewood as well as recreating a simple virtual fire (which I'd love to try and animate later), made of transparent flame textures and two simple particle systems representing flying embers. 

* 336 pictures on a Pocophone F1.
* Processed with Colmap, OpenMVS, Blender and BakeMyScan.
* Fire crackling sound from [Freesound.org](https://freesound.org).
* The sticks/logs come from [this collection](https://skfb.ly/6JqUw).

Rustic Fireplace

![](http://an-uhelgoad.franceserv.com/morh4.jpg)

Located in the [Huelgoat forest](https://goo.gl/maps/uJw11Zwyqr22), the sibling of the legendary Brocéliande forest, this site (called the "Boar Pond") was a safe haven for the animals which populated the forest.

It is even rumored that gold flakes can be found at the bottom of the pond - explaining its orange color - leaking from King Arthur's treasure, which Merlin the Enchanter is said to have hidden nearby...

Although the scanning process went quite fine, I was not really glad of the reconstruction result due to constantly changing lighting conditions... But I compensated by trying to recreate a satisfying water surface with Sketchfab's materials and effects, and actually ended up quite happy with the final model !

* 441 pictures (Pocophone F1).
* Colmap, OpenMVS, Blender and BakeMyScan.
* [Background sound](https://freesound.org/people/Mjeno/sounds/405133/) from Mjeno on [FreeSound.org](https://freesound.org)

"La Mare aux Sangliers"

**Licensed under CC-Attribution**: citing is cool!

*Hint: Follow the righteous path of annotations* ;)

Scanned and processed with Open Source software:

* scan with Colmap and OpenMVS
* post-process within blender and [BakeMyScan](http://bakemyscan.org) (4.2M -> 15k tris)
* texture correction with GIMP

Thumbs up to Philippe, the owner of my neighboorhood antiques shop ([La Brocante du Phare](http://goo.gl/szA4Sa) in Roscoff), who agreed to lend me this statuette but did not have much info: It is a buddha, is not made out of Jade, is heavy and about 25cm tall ;)

![](https://i.imgur.com/Ok55RcK.jpg)

My scanning "rig": Canon EOS 1100D (one of the cheapest DSLR 6 years ago), cheap Amazon lens-flash, Ikea furniture. And yes, I used my paper bin !

![](https://i.imgur.com/OIXJMgf.jpg)

Colmap gave better results than Photoscan... 146 pictures.

![](https://i.imgur.com/1tGa9rc.jpg)

I had to manually carve (and smooth) the space between the arm and the body.

![](https://i.imgur.com/JNVawgD.jpg)

Processing...

Open Source Buddha

Photogrammetry reconstruction of a shell I had laying around, of which sadly I don't even know the origin...

Processed from 110 pictures from my now usual Colmap/OpenMVS/Blender/BakeMyScan workflow.

If you are curious about what's inside, Peter Fried from NYU ( @ARC-3D ) has uploaded an interesting version of such a shell, mixing photogrammetry and CT-scan, and viewable [here](https://skfb.ly/68VBV).

Murex Shell

This Bronze cast created in 1927 by François Rudier is part of the [monumental stairs of Marseille Saint Charles](https://goo.gl/maps/EjkGZ38qHaGS66iB7), in France.

The statue - called "La Chasse" (hunting) - is part of a group of six similar pieces representing the region's products (other being fruits, flowers, fishing, cereal harvesting and grapevine harvesting).

Photogrammetry was done from 164 pictures (Pocophone F1) and processed with Colmap and OpenMVS, but I had to correct some of the upper parts of the sculpture in Blender as I did not have a way to get high angles when shooting.

![](https://www.fortunapost.com/3556/3545.jpg)

"La Chasse"

![](https://i.imgur.com/jSTrKfy.jpg)

Located on the shore of [Plounéour-Trez](https://goo.gl/maps/j3MMCNg8Lzo), this cross from the XIXth century faces the Bay of Goulven - an ideal location for water sports or on-foot shell fishing depending on the tide - and is a reminder of the past strong religious culture of Brittany (a region of France).

In Breton, "Beg" means headland, and "Groaz" translates to cross.

* 542 pictures (Pocophone F1).
* Colmap, OpenMVS, Blender and BakeMyScan.
* [Background sound](https://freesound.org/people/suukadi/sounds/410323/) from Mjeno on [FreeSound.org](https://freesound.org)

"Beg Ar Groaz"

**About the sculpture**

![](https://i.imgur.com/ig60lLs.jpg)

This sculpture, created in 1989 by Patrig Ar Goarnig and Lucien Crenn, is located near the cove of "Porz Ar Streat" in Plouescat, in France.

For those of you who prefer Google Maps, you'll find it [here](https://goo.gl/maps/ESzuX8LEXkx) ;)

It represents a Seahorse, which tail is wrapped around a sleeping cat on one side, and a Yin-Yang symbol on the other.

**Process**

![](https://i.imgur.com/HxxJHR4.jpg)

The scan was made from 140 pictures (the last of them under a heavy rain), reconstructed with Colmap and OpenMVS and optimized to a lowpoly version (10k triangles) with Blender and [BakeMyScan](http://bakemyscan.org).

I also used a weight-painted metric for the retopology to keep the important details: do not hesitate to check the wireframe view.

Finally, and due to parts of the statue being wet, I also decided to ditch the computed albedo map and used a mix of granite PBR textures instead.

Granite Seahorse

**3D scan of an ASICS running sneaker reconstructed with AliceVision/Meshroom 2025.1.0.**

**Pipeline:** single iPhone 14 Pro 1080p video (orbit, 30 fps, ~50s) → frames @ 4 fps → SAM3 (text prompt "sneaker") for background segmentation → COLMAP-style sparse + dense reconstruction via Meshroom's full `photogrammetry` template → **3D Delaunay tetrahedralization + Graph Cut Max-Flow surface extraction** (instead of Poisson) → native Texturing node bakes a 4096×4096 diffuse map from the masked frames.

**Stats:** 203 frames, 85,730 vertices, 133,521 triangles. Full pipeline ran on a GCE L4 GPU in ~115 minutes.

Sneaker_meshroom_delaunay_sketchfab

Pipeline (reproducing obotx/digital-twinning Path 1): extract frames @ 4 fps → SAM3 text-prompt mask ("cup") → COLMAP feature_extractor on full images (exhaustive + relaxed mapper thresholds: 251/252 frames, 53k sparse points) → image_undistorter on black-bg → pixel swap [0..30]→[255,255,255] → PINHOLE sparse → patch_match_stereo + geom_consistency (CUDA, A100) → stereo_fusion (793,296 dense points) → Poisson mesher (open3d depth=10, density-trim 5%, LCC) → Taubin smoothing.

 Result: ~793k dense points → ~150k mesh faces / 25.8 MB.

Classical photogrammetry baseline. Cup ornament band (green rim + blue pattern) reconstructs OK because of texture, but the smooth white body produces few stereo matches — Poisson then collapses the sparse point cloud into a flat "puddle" with one elongated wing. Illustrates the known limit of SfM+MVS on glossy/textureless objects vs Gaussian-based methods.

  Reference: github.com/obotx/digital-twinning.

Mesh_cup_colmap_poisson

Pipeline: SAM3 text-prompt masking → COLMAP sparse (sequential matcher,
  252 frames) → MeshSplat training (A100, 30k iter, surface regularizers
  + indoor flag, ratio 0.95).

  Raw output: 213,785 v / 414,334 f, ~198,000 disconnected components — "forest of needles".
  
  Regularization: bbox crop, statistical outlier removal, voxel downsample
  (5mm), Poisson reconstruction (depth=10, 10% density trim), largest
  connected component (98.3%), quadric decimation to 20k faces, Taubin
  smoothing (30 iter), vertex color transfer.

  Final: 11,009 v / 20,000 f / 0.82 MB. 95% vertex reduction.

  Code: github.com/meshsplatting/mesh-splatting (CVPR 2026).

Mesh_meshsplat_ultra_a100_v2

**3D scan of a mechanical keyboard reconstructed with AliceVision/Meshroom 2025.1.0.**

**Pipeline:** single iPhone 14 Pro 1080p video (orbit, 30 fps, ~60s) → frames @ 4 fps → SAM3 (text prompt "keyboard") for background segmentation → COLMAP-style sparse + dense reconstruction via Meshroom's full `photogrammetry` template → **3D Delaunay tetrahedralization + Graph Cut Max-Flow surface extraction** (instead of Poisson) → native Texturing node bakes a 4096×4096 diffuse map from the masked frames.

**Stats:** 81 frames, 101,432 vertices, 147,881 triangles. Full pipeline ran on a GCE L4 GPU in ~50 minutes.

Keyboard_meshroom_delaunay_sketchfab

Pipeline: extract frames @ 4 fps (q=2 JPG) → DAv2 depth maps →Metric3D normals → COLMAP sparse (exhaustive matcher + relaxed mapper thresholds: 252/252 frames registered, 53k points, 0.53 px eprojection error) → image_undistorter → SAM3 text-prompt segmentation ("cup") → MeshSplat training (A100 40GB, 30k iter, surface regularizers active via DAv2 + Metric3D priors, indoor flag, ratio_threshold 0.90) → 5-step segmentation extraction (extract_images, segment, run_single_object, create_ply).

Result: 105,771 vertices / 143,457 faces / 3.4 MB.

Raw MeshSplat output, no post-processing. Cup body and ornamentclearly recognizable (green rim, blue medallion pattern, white base). Surrounding shrapnel triangles and table fragments are the per-Gaussian rasterizer artifact — MeshSplat is a rendering- first method, output is structurally a point cloud where each Gaussian becomes one disconnected triangle.

  Code: github.com/meshsplatting/mesh-splatting (CVPR 2026).

Mesh_meshsplat_cup

Pipeline: SAM3 text-prompt masking → COLMAP sparse (sequential matcher, 252 frames) → MeshSplat training (A100, 30k iter, surface regularizers+ indoor flag, ratio 0.95).

  Raw output: 213,785 v / 414,334 f, ~198,000 disconnected components — "forest of needles".
  
  Regularization (closed-surface variant): bbox crop, statistical outlier
  removal, voxel downsample (5mm), Poisson reconstruction (depth=10,
  density trim 2% — less aggressive than v2), largest connected component
  (99%), quadric decimation to 30k faces, Taubin smoothing (20 iter),
  vertex color transfer.

  Final: 15,137 v / 30,000 f / 1.16 MB. 93% vertex reduction.

  Code: github.com/meshsplatting/mesh-splatting (CVPR 2026).

Mesh_meshsplat_ultra_a100_v3

Another batch of scanned sticks !

The sticks are in .fbx format, and each one is made of 150 triangles, and comes with 256px albedo and normal maps.

If you need high resolution of those props or different file formats, you'll probably find what you're looking for on the [Store version of this model](https://sketchfab.com/models/448b03c40ace4833abc15f928905c86c), which comes with 5 different LODs for each model, the higher ones having 15k triangles and 4096px textures.

I also made a model made of more than 2000 instances of those assets to have an idea of the possible WebGL performance: ["That's one big pile of stick"](https://sketchfab.com/models/ded075151a294ad094b43903ba2f8f59).

Licensed under **CC-attribution**: feel free to re-use them with the appropriate credit, and do not hesitate to drop a comment if you used them for your own creations, I’d be happy to know !

Created with love and Open Source tools only: Colmap, OpenMVS, Blender and [BakeMyScan](http://bakemyscan.org).

Lowpoly Sticks - Free Download

A bunch of rusty tools, scanned directly from my toolbox, and optimized to lowpoly PBR props with Open Source tools only: Colmap, OpenMVS, Blender and [BakeMyScan](http://bakemyscan.org) !

Should you need them, all the LODs you might ever dream of for this collection are [available on the Store](https://skfb.ly/6FSYn):

* 15k tris - 4096px
* 5k tris - 2048px
* 1.5k tris - 1024px
* 500 tris - 512px (the ones you see here!)
* 150 tris - 256px

These assets are especially optimized for realtime applications: VR, AR, WebGL, game engines, mobile apps, kitbashing... Here is an example of particle instancing with those props: ["The Toolball"](https://skfb.ly/6FSYs).

Licensed under **CC-attribution**: feel free to re-use them with the appropriate credit, and do not hesitate to drop a comment if you used them for your own creations, I'd be happy to know. 

And if you enjoyed those assets, you might find other interesting stuff in [this lowpoly collection](https://skfb.ly/6FSYx) ;)

Lowpoly Tools - Free Download

**Pipeline:** single iPhone 14 Pro 1080p video (orbit, 30 fps, ~63s) → frames @ 4 fps → SAM3 (text prompt "cup") for background segmentation → COLMAP-style sparse + dense reconstruction via Meshroom's full `photogrammetry` template → **3D Delaunay tetrahedralization + Graph Cut Max-Flow surface extraction** (instead of Poisson) → native Texturing node bakes a 4096×4096 diffuse map from the masked frames.
**Stats:** 252 frames, ~94k vertices, ~118k triangles. Full pipeline ran on a GCE L4 GPU in ~90 minutes.

Asset Library

About These Assets