Here’s the original post, and links to all posts
I have previously outlined my goal of testing multiple photogrammetry solutions on a single dataset, and reporting times and results.
I’m using a dataset based on photographs of this Styracosaurus model (I’ve had it since I was quite young):
The dataset has 53 photos in total, and is available from this link. [This will be moved to figshare in due course].
The model is about 12 cm in total length, has texture for the scales, and a reflective brass nameplate on the base. The model was situated on a glass desk, and there may be reflection issues from that.
In this post, I’ve generated models using Photoscan at both low and high settings (i.e. for alignment, dense reconstruction, and mesh generation settings were set to either all ‘low’ or all ‘high’. Textures were generated at 4096px for both settings.
I was using version 1.2.4 of Photoscan.
On low settings, timings were as follows:
Alignment: 48.3 seconds
Dense Reconstruction: 66.37 seconds
Mesh Reconstruction: 20.38 seconds (19999 faces)
Texturing: 140.8 seconds
Total time: 275.85 seconds
You can view and download the model on sketchfab and here:
Note that if you change the rendering mode to ‘matcap’ you can view the mesh without texture
For high settings, I ran a batch process which meant it was difficult to parse out how long each stage took, and I only have the following:
Alignment: 434.01 seconds
Dense Reconstruction: 1556.7 seconds
Mesh Reconstruction: unknown (1782382 faces)
Total time: 2706.48
Here are a couple of images of the stages:
And as above, here’s the final model (please note, I’ve no idea how much sketchfab may or may not downsample meshes – I will upload all originals to figshare when I’m done testing). You can see that the surface is very rough:
I also captured system usage during processing of the ‘high’ dense reconstruction:
Agisoft Photoscan – Summary
I’ve started with perhaps the most predictable piece of software, Photoscan. The reconstruction was pretty robust, though both models general suffered from a lack of detail at the model’s rear left. The reflective brass name plate didn’t throw up any issues, which I thought it might.
Both models were quite noisey (I haven’t done any manual cleaning) even with ‘aggressive’ noise reduction set. The ‘high’ model had a very rough surface, which I attribute to noise. There are some issues with the dataset, particularly regarding depth of field in some shots, but the result still has quite a bit of noise in there.
Of course, the whole process was incredibly easy to do, and running the ‘high’ settings through a batch process meant I could go have dinner and come back to a finished model.
As we don’t have anything to compare Photoscan to, I’ll leave the summary there for now and come back to it in a round-up post.