{"id":198,"date":"2026-02-27T22:29:53","date_gmt":"2026-02-27T22:29:53","guid":{"rendered":"https:\/\/projects.etc.cmu.edu\/dots-to-depth-studios\/?page_id=198"},"modified":"2026-03-23T17:35:08","modified_gmt":"2026-03-23T17:35:08","slug":"week-7-blog","status":"publish","type":"page","link":"https:\/\/projects.etc.cmu.edu\/dots-to-depth-studios\/week-7-blog\/","title":{"rendered":"Week 7 Blog"},"content":{"rendered":"\n

This week is all about preparing for Halves \u2014 our mid-semester presentation that, as our faculty advisor John Dessler aptly put it, is really more for ourselves than for faculty. <\/p>\n\n\n\n

\n
\n
\"\"
Rehearsing for the Halves presentation in ETC’s RPIS space<\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n
\"\"
Grace took a photo of our rig to include in her slides<\/figcaption><\/figure>\n<\/div>\n<\/div>\n\n\n\n

Putting this presentation together has \u201cforced\u201d us to pause, reflect on what we\u2019ve actually learned, and define a smart direction for the rest of the semester. The key words here are learned<\/em> and smart<\/em>.<\/p>\n\n\n\n

Unlike a typical ETC project that primarily showcases what has been built, our team has been operating with two parallel prongs from day one: research AND production.<\/p>\n\n\n\n

It\u2019s rare to work with a technique so new that even industry professionals are still figuring out what to do with it. There\u2019s considerable buzz around 3D Gaussian Splatting (3DGS) in the visual effects sector, yet even top-tier studios with advanced equipment and substantial funding are still exploring how to best leverage it. We\u2019re contributing to that same exploration. Our goal is to make a real-world contribution to the evolving conversation around 3DGS \u2014 and that\u2019s something we\u2019re genuinely proud of.<\/p>\n\n\n\n

As for being smart<\/em>: we don\u2019t have unlimited time or funding. With six weeks left, we need to define a final deliverable now \u2014 something tangible and compelling. Not just a product for the sake of having one, but a piece that thoughtfully showcases what we\u2019ve discovered in a cohesive and presentable way.<\/p>\n\n\n\n

So what have we learned so far?<\/p>\n\n\n\n

1. 3DGS does not behave like a mesh model \u2014 and treating it like one defeats its purpose.<\/strong>
For instance, we attempted to place a 3DGS-scanned object onto a dark, snowy street at night. Because the object wasn\u2019t originally captured under those conditions, it felt visually disconnected from the environment. It casts no true shadow, and we cannot edit its \u201csurface\u201d to accumulate snow \u2014 point clouds have neither physical surfaces nor traditional shading behavior. Technically, we could build a low-poly proxy mesh to fake shadows and create colliders for snow interaction. But that additional manual labor undermines one of 3DGS\u2019s key advantages: reducing modeling time and manpower.<\/p>\n\n\n\n

\n
\n
\"\"
Digitally modeled snow scene<\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n
\"\"
Adding 3DGS models: CMU’s CUC building in the background, and a statue in the foreground<\/figcaption><\/figure>\n<\/div>\n<\/div>\n\n\n\n
\n
\n
\"\"
Adding shadows to the foreground 3DGS models<\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n
\"\"
Comparison experiment: the same building and statue rendered (as mesh models) using MeshyAI<\/figcaption><\/figure>\n<\/div>\n<\/div>\n\n\n\n
\n
\n
\"\"
Side-by-side view: 3DGS model (left), MeshyAI model (right)<\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n
\"\"
Low-poly proxy used to create a shadow for the 3DGS model <\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n
\"\"
Result after placing the proxy inside the 3DGS model <\/figcaption><\/figure>\n<\/div>\n<\/div>\n\n\n\n

2. A particle-based medium is powerful on its own terms.<\/strong>
Rather than forcing points to behave like meshes, we began asking: what are points naturally good at? Think of pointillism in fine art \u2014 the beauty comes from embracing the medium\u2019s inherent qualities. Points disperse, gather, shimmer, dissolve, and flow. They excel at fluidity, transformation, and atmosphere. When we stop trying to forcefully control the 3DGS models \u2014 to edit, rig, or collide them like rigid geometry \u2014 new creative possibilities emerge.<\/p>\n\n\n\n

\n
\n
\n