We have been asked about the recent launch of object capture, a photogrammetry based workflow, claiming to have solved 2D to 3D problem. Photogrammetry has been around for several years and has undergone a lot of advancements. Meshroom, Micmac, and several other software that have shown incredible results. So, there has been no significant technology advancement in the object capture release except that it has become accessible to the high-end laptops. 

We tested the object capture on a variety of objects ranging from simple objects such as tubes to shoes. First, you need to take lots of images from various angles. Second, the software works only on the latest laptop or the older ones with AMD GPU. 

  1. Taking Images: You need a proper well-lit set-up to take multiple pictures.

       2.Once you run the program, it will deliver a 3D model like this:

Photogrammetry result for tube image

Even after you take 100s of images, the API struggles to generate high quality 3D model. Photogrammetry doesn’t work for smaller shiny surfaces. All photogrammetry technologies work really well on non-reflective objects which would still require expert involvement to get to a quality that is indistinguishable from reality.

Lastly, we tried this on non-reflective surfaces with over 300 images of shoes. The results look like this:

In most cases, an expert would be required to carefully capture the model and clean the model afterwards to get a quality that can be used in AR/VR/ Photos/ Videos. 

If you are looking for exceptionally high quality without expert intervention, we see procedural workflow combined with computer vision as the best approach to achieve quality that will be indistinguishable from reality and does not require any expert software. Procedural is a way in which each characteristic of the model is defined and can be controlled by anyone. E.g A shoe has a sole, upper layer, laces, and heels. If all parameters of shoes are defined, you can achieve a hyper-real shoe that cannot be distinguished separately from eyes. Moreover, it allows anyone (non-expert) to control all parameters. Procedural will struggle for objects such as statues and highly artistic objects but will provide an exceptional quality when it comes to our daily products.  

In case of use-case for photo-shoot of eCommerce items, we fail to see a non-expert workflow in photogrammetry that will provide you with a very clean 3D model which can be used to create photos that do not require high-end camera and photoshop. Moreover, the time taken to capture and access to expensive laptops, means restricted use. 

We have created a non-expert workflow for consumer good items on the web which truly democratises a subset of 3D modelling. We will take it to all eCommerce products to allow photography to become completely virtual. You can see the demo below of non-expert workflow for eCommerce consumer products. 

Feel free to contact us at contact@zeg.ai to learn more about non-expert 3D modelling.