← All examples

Cycladic figurine: from 72 photos to a measured mesh

How to process the cycladic figurine dataset. A step by step guie in KIPES: tie points, camera orientation, scale, self-calibration, then a dense point cloud and a mesh in millimetres.

The dataset: 72 photos in one zip, 240 MB

Downloadfrom doi.org

The dataset

Chad Everett Hutchens and Charles Koenig published 72 photos of a replica Greek Cycladic figurine for a lesson on 3D digitization, under CC BY 4.0. A Canon EOS 5D Mark III with a 50 mm lens took them in three rings of 24 around a turntable, each 5760 × 3840 pixels. Two printed scale cards lie beside the figurine. Each card carries two circular targets, and the distance between them is written on the card: 0.05005 m and 0.05010 m.

Download the zip (240 MB) from the dataset's page and unpack it: https://doi.org/10.15786/16543554

A small white figurine on a white turntable, two grey scale cards with black-and-white circular targets beside it, black background.
One of the 72 photos (photo 10). The cards read 0.05005 and 0.05010.

1. Create a space and a subspace

On the first start, KIPES opens Set up your workspace. Type a space name and a subspace name (here Examples and Cycladic figurine) and press Create. Later spaces and subspaces are added with the + buttons in the sidebar.

The form "Set up your workspace" with the space name Examples and the subspace name Cycladic figurine.
The first start asks for a space and a subspace.

2. Import the photos

On the Workspace tab, the import button of the Images card opens the file dialog: select all 72 JPGs. KIPES reads the camera from the files and asks once for its sensor. Keep the name Canon EOS 5D Mark III, choose Full Frame (36 × 24) as the sensor size and The whole import, then press Apply.

The dialog "Which sensor do these images belong to?" with the name Canon EOS 5D Mark III, the sensor size Full Frame (36 × 24) and The whole import.
The sensor is asked once for the whole import.

3. Detect tie points and the card targets

Open Target Detection.

  1. Choose the target type Precise under Natural features, set Keypoints per image to 50000 and start the detection for All images. KIPES finds natural features in every photo and links the same points across the photos.
  2. Choose Quadrant circle and start All images again. Confirm the question with the red Delete button and leave Also delete … other types unticked: nothing is deleted then. KIPES finds the four card targets in the photos.
A photo of the figurine and the two scale cards, the cards' circular targets outlined in blue.
Tie points and card targets on all 72 photos.

4. Orient the cameras

Open Relative orientation and press Compute with the settings as they are. All 72 cameras take their places in three rings around the figurine; the camera's focal length and its main lens distortion are estimated on the way.

72 numbered cameras in three rings around the figurine, seen from above, the figurine's points in yellow.
All 72 cameras oriented in three rings.

5. Give the network its scale

  1. Press Resect: the four card targets get point numbers.
  2. Look up which numbers sit on which card; Target Detection shows them on the photos. In our run, 10014427 and 10014428 lie on the 0.05005 card, 10014426 and 10014429 on the 0.05010 card.
  3. Under Constraints, add a Scale bar for each card: its two points, the length in millimetres (50.05 and 50.10) and the status Fixed.
  4. Press Compute again and confirm Overwrite. The network is now in millimetres; both cards fit it to 0.01 mm.
A close-up of the scale card labelled 0.05005 with its two circular targets numbered 10014428 and 10014427.
The point numbers of the 0.05005 card.
The Constraints table with two scale bars: Card 0.05005 from 10014427 to 10014428, 50.05 mm, and Card 0.05010 from 10014426 to 10014429, 50.10 mm, both Fixed.
The two cards as scale bars.

6. Calibrate the camera in the bundle adjustment

A new sensor's lens parameters are set to Free (see the Sensors tab), so the bundle adjustment calibrates the lens together with the network. Open Bundle Adjustment, add the two cards under Constraints again, this time as Stochastic scale bars with σL = 0.01 mm, and press Compute.

In our run all 72 cameras and 8 623 points fit with a reprojection error of 0.77 pixels, and both cards agree with the network to 0.024 mm.

The bundle adjustment's constraints table with the two cards as Stochastic scale bars with σL 0.01.
The bundle adjustment with the cards as weighted scale bars.

7. Dense point cloud and mesh

Open 3D Reconstruction and set:

  • Sensor: the bundle adjustment
  • Point spacing: 0.25 mm
  • Multi-baseline pairs, Mesh and Surface fusion: on

Leave the other options as they are and press Compute. KIPES matches 144 photo pairs and fuses their depths into one surface.

The 3D Reconstruction settings: Sensor Bundle adjustment, Point spacing 0.25 mm, Multi-baseline pairs, Mesh and Surface fusion ticked.
The settings for the point cloud and the mesh.

The result

About 790 000 points and 1.5 million triangles, in millimetres. The figurine keeps its folded arms, the waist line and its base; the print, the logos and the targets of both cards can be read on the surface. Where few photos look at the surface, at the top of the head and on the upper back, small holes remain.

The coloured mesh of the figurine standing between the two scale cards, with the rim of the turntable as a thin ring.
The mesh: figurine, cards and the turntable's rim. The small yellow dots are tie points; the thin lines rising from the four card targets are their normals.
A closer view of the coloured mesh: the figurine with folded arms, the cards' print and targets readable.
Closer: arms, waist line and the cards' print.