Exercise: Process Your Own Dataset on SfM

Laser Scanning vs. Photogrammetry (click to enlarge the figure)

In this exercise you will have the opportunity to practice SfM processing using Agisoft Metashape. After completing this exercise you should be able to: 

• Process a dataset that you have captured yourself
• Reflect on how capturing strategies, mistakes, and omissions are reflected on the processing stage.

Here you will follow the same workflow as described in the 'Processing with Metashape' lesson. Before you start this exercise, it is advisable to capture your own dataset, even by using your smartphone camera, rather than using a dataset captured by somebody else. In such a way, you will be able to better observe how capturing is reflected in processing. Since this is just a trial, it would be good to limit the number of images to less than 50, so the processing gets done quicker. Otherwise, you can download any dataset from captured by Daniel Pett, Head of Digital and IT at Fitzwilliam Museum, UK, from the following GDrive folder: https://drive.google.com/drive/folders/1LgHMXdZYsyGWj-Q-Qm4LmuSnfrcSghNE?usp=sharing

Your task is to:

1. Manually go through your images and divide them in different folders (duplicates, blurry, TIFF, RAW, JPG etc.)
2. Add the images to Metashape and estimate their image quality
3. Disable (but do not delete) the images that are below the quality threshold
4. Create masks for all your images leaving only the main object of interest unmasked
5. Align Photos (use quality settings that can be handled by your device - for this stage an average laptop can cope with high quality) and make sure to select 'Apply masks to'
6. Once this stage is done observe your sparse point cloud and see if there are any major gaps. Why is that? Are there any points outside the area of interest? If so, use the selection tools and delete them
7. Build a Dense Cloud (use quality settings that can be handled by your device - this is the most computationally intensive stage, therefore you should try with medium first. If your device is powerful enough, you should select high). Depending on the number of photos you are using and the selected quality, this stage might take from a few minutes to several hours.
8. Duplicate the Dense Cloud
9. Build a Mesh using the default settings
10. Use the tools in the menu to a) close holes; b) smooth mesh; c) decimate mesh. What does each of these do? In which cases they may be helpful?
11. Calibrate the colours of your model using the 'model' as the source data. Do you see any difference? Also, try the 'white balance' calibration. 
12. Build the Texture using the default 'Generic' mode
13. Export the model in .obj and .stl formats. What are the differences between the two?
14. Open the models in your 3D viewer. If you don't have one you can download the free/open-source software MeshLab. Do you observe any differences between the way the model was displayed in Metashape and in MeshLab?
15. Generate a processing report
16. Write a short blog post or a text for your reflective diary (c. 300 words) to summarise the processing of the dataset and explain what you might have changed (or tried out) if you had the chance to capture the dataset again. What are the challenges you faced (or you could potentially face) in processing? How does capturing impact the processing?