The capturing phase for the second case study in the NUIM-AFF621: Remaking the Physical 3D – recording assignment was discussed in the last post. This post is concerned with the processing of the captured data, being 124 photographs of a statue in the outdoors. In the first case study, I discussed the difficulties I had in the processing phase using Agisoft PhotoScan Professional, however, I also concluded that this was as a consequence of the captured data, which was not “Good”. Similar to the previous attempt, I used the same laptop, followed the same workflow, and referred to the Agisoft PhotoScan Professional guide for beginners “3D Model Reconstruction” to process the photographs of the statue.
I uploaded the images and checked for the image quality, and approximately 10 photographs had the image quality from 0.45 to 0.5, so I deleted them from the programme. Then, I began the task of masking the images. This was extremely time consuming as I had to use the scissor tool for each one as the magic wand tool was not appropriate due to the amount of detail in the background. This was also difficult due to the colour of the statue, as shown below. Indeed, it took approximately 7-8 hours to complete 114 photographs. Nonetheless, I think it was worth it in terms of the final result.
Unlike the first case study where I had difficulties in the photo alignment process, this time every photo aligned and it was just a matter of creating the dense point cloud. This took some time as I opted for a tie point of 2000 after reading some comments on the Agisoft Community Forum (tie points). Once this finished I used the clean-up tool to get rid of any stray points and I submitted to create the final mesh.
Previously, I used Verold to make the 3D model available online from an obj. file exported from Agisoft PhotoScan, so this time I used Sketchfab. It was relatively easy to upload the obj. file, did not take long to process, and was easy to embed. Indeed, I am quite proud with how this turned out, and glad that I put in the effort with the capturing and the masking of the photographs. I would certainly use Agisoft PhotoScan Professional again, but realise how important the capturing phase is for this type of endeavor.
On the other hand, working with 123D Catch on this case study proved to be less straightforward. In Post 5, I demonstrated how 123D Catch from Autodesk was used as a processing application to create a 3D model for the the first case study. The post demonstrated issues I had related to the captured data, and also in my approach. Additionally, I highlighted that the 123D Catch application seemed to have an error in processing a capture scene from the manual stitching process. However, working with 123D Catch in processing the photographs for the statue was problematic on a different front. I followed the advice given (by me) for using 123D Catch from the last case study, and went through a workflow to carefully choose 62 images from the master folder of equal distance and spacing, I uploaded the images, and created a new capture. In the model scene, every photograph auto-stitched so there was no need for manual stitching, I clicked on 2-sided model, and merely had to clean up the unwanted background. However, when I tried to use the clean up tool to get rid of the unwanted background, the programme kept reporting a non-response each time I highlighted an area to delete. This went on for some time, until the programme finally crashed and closed. I tried it again a few hours later, but the programme was still on a go-slow, even though I had no other applications running, and my broadband connection was strong.
My first instinct was to give up, as I had spent an enormous amount of time in fulfilling this assignment, and while doing so, I had very little time, or computer capacity to pursue other tasks. However, I did return for another attempt. Using 55 photographs with good spacing and distance, I created a new capture, and each photograph auto-stitched in the opening model scene. I then followed (step-by-step) a 123D Catch tutorial on correcting Mesh Details in the model scene. Nonetheless, the programme was consistently non-responsive.
At this point, I decided not to pursue the reasons for my latest conundrum with 123D Catch. I did manage to process a scene that had some of the background excluded before the programme crashed, and that is shown below. To conclude, in watching many tutorials on 123D Catch from Autodesk, it looks like a very impressive and user-friendly application for creating a 3D model, in comparison to the efforts required for Agisoft PhotoScan Professional. However, from my experience in this assignment, I did not find 123D Catch as user-friendly in terms of how it is portrayed in the tutorials, and I would easily prefer to use Agisoft PhotoScan Professional for any future endeavours.