CS491 / CS492 Senior Design Project · Bilkent University

AssetForge

Extracting editable, game-ready 3D models from a single photo.

About the Project

The name joins asset, the models and art that make up a game, with forge, a workshop where raw material is shaped into something usable. Our tool takes a single photo or concept drawing and turns it into a 3D model built from separate, editable parts that you can use directly in a game engine.

The Problem

Small game studios and indie developers need many 3D models. Building them by hand in tools like Blender takes time and experience, and hiring a 3D artist is often beyond a small team's budget.

Recent AI tools such as Meta's SAM 3D and Meshy generate a model from one photo in minutes, but the result is usually a single, irregular mesh. Enlarging only a table's top while leaving its legs alone means cleaning up or remodeling everything by hand.

Our Approach

We build on 3-Sweep (Chen et al., SIGGRAPH Asia 2013). For each part of an object, the user sketches a cross-section and then its sweep axis, and the system lifts that part into 3D. Because the user decides how the object is split up, the result is a set of independently editable parts.

We keep this sketch-based workflow and add AI where manual work is hardest, such as selecting the object in the photo or completing the hidden back side.

Photo Object selection User sketch 3D parts Completion & merging Export (Unity, Godot)

Planned pipeline. Dashed boxes are steps where we plan to use AI assistance.

Features Under Consideration

We will add at least three new features on top of the base method. The final selection will be made with our supervisor after the core method is implemented. Current front-runners: hidden-surface completion, depth estimation, and game-ready export.

AI Assistance

Automatic object boundaries with a Segment Anything–style model; suggested part type, axis, symmetry and initial depth; back-side texture inpainting; monocular depth for non-geometric shapes; region-limited, text-prompted edits.

Richer Geometry

Free-form B-spline cross-sections beyond circles and rectangles; editable combinations of sweeps, primitives and extrusions; hinges between parts for simple automatic rigging.

Game-Ready Output

Merging overlapping parts into one watertight mesh, polygon reduction (LOD), collider generation, and export to Unity, Godot or Blender.

Target users: small game studios, indie developers and students learning game development. We plan to distribute the tool as a Unity or Godot plugin.

Team

Department of Computer Engineering, İhsan Doğramacı Bilkent University.

Can Akpınar
Student ID: 22103263
[email protected]
Defne Lal Oruncak
Student ID: 22201809
[email protected]
Leyla Özpınar
Student ID: 22202030
[email protected]
Fevzi Yarkın Müftüoğlu
Student ID: 22102577
[email protected]

Supervisor, Instructors & Innovation Expert

Supervisor
Yusuf Sahillioğlu
[email protected]
Instructor
Mert Bıçakçı
[email protected]
Instructor
Hüseyin Açacak
[email protected]
Innovation Expert
Cem Çimenbiçer

Reports & Deliverables

All project reports and presentations will be published here as they are submitted.

DeliverableCourseDueStatus
Project SummaryCS491Oct 20, 2026Submitted
Project Specification ReportCS491Nov 20, 2026Coming soon
Analysis & Requirements ReportCS491Dec 11, 2026Coming soon
End-of-Semester Presentation & DemoCS491Dec 2026Coming soon
Detailed Design ReportCS492Spring 2027Coming soon
Final ReportCS492Spring 2027Coming soon
Final Presentation & CS FairCS492Spring 2027Coming soon

To publish a report, place the PDF in a reports/ folder and turn its name into a link, e.g. <a href="reports/specification.pdf">.

Timeline

Planned milestones over two semesters.

  1. October – November 2026
    Implement the core 3-Sweep method and select features. Project Specification Report on Nov 20.
  2. December 2026
    Analysis & Requirements Report (Dec 11), end-of-semester presentation and demo.
  3. Spring 2027 (CS492)
    Develop selected features, test the tool with game developers, present at CS Fair.

References

  1. T. Chen, Z. Zhu, A. Shamir, S.-M. Hu, D. Cohen-Or. “3-Sweep: Extracting Editable Objects from a Single Photo.” ACM Transactions on Graphics 32(6), SIGGRAPH Asia 2013.
  2. Meta AI, SAM 3D announcement, 2025. ai.meta.com/blog/sam-3d