SmartRoom3D: Intelligent 3D Room Design
(Proposal for SHREC 2025)
Organizers:
- Trong-Thuan Nguyen*, University of Science, VNU-HCM, Vietnam, ntthuan@selab.hcmus.edu.vn
- Viet-Tham Huynh, University of Science, VNU-HCM, Vietnam, hvtham@selab.hcmus.edu.vn
- Minh-Triet Tran*, University of Science, VNU-HCM, Vietnam, tmtriet@hcmus.edu.vn
- Tam V. Nguyen, University of Dayton, U.S.A, tamnguyen@udayton.edu
…
1. Motivation
Interior design rapidly evolves with 3D analysis, computer vision, and natural language processing advancements. The 3D-FRONT dataset, comprising 8,797 fully furnished rooms and 7,302 high-quality furniture models, presents a unique opportunity to develop an intelligent system that integrates spatial reasoning, visual aesthetics, and language understanding.
This challenge aims to foster research in multi-modal retrieval, enabling systems to suggest context-aware interior design solutions that balance both functionality and style.
2. Challenge Statement
Objective
Develop an integrated, multi-modal retrieval system capable of performing the following task:
Task: Intelligent Furniture Retrieval
Participants will design a system that retrieves and ranks furniture suggestions based on both spatial configurations and natural language descriptions.
3. Dataset
We will leverage the 3D-FRONT dataset, which includes:
8,797 fully furnished 3D room models
7,302 high-resolution furniture models with textures
This dataset provides spatial relationships, material properties, and aesthetic styles, enabling a comprehensive evaluation of retrieval techniques.
To support participants, we tentatively plan to provide:
100-200 queries in the training set for reference.
100-200 queries in the test set to evaluate system performance.
4. Task Definition
Objective
The goal is to develop a system that automatically suggests furniture items that complement an existing room design, considering both spatial positioning and design constraints specified in natural language.
Inputs
Spatial Input:
A set of 3D coordinates defining a specific placement region within the room.
Language Input:
A descriptive sentence specifying style preferences, functional requirements, and constraints.
Example: "A modern sofa for a compact living room with minimalist decor."
Outputs
Ranked Furniture List:
A ranked list of furniture items based on relevance to the given spatial and linguistic criteria.
5. Evaluation Metrics
To assess the quality and effectiveness of retrieval systems, we propose the following metrics:
Precision & Recall: Measure how accurately retrieved furniture items match the spatial and linguistic criteria.
Mean Average Precision (mAP): Evaluate ranking quality across multiple queries.
Normalized Discounted Cumulative Gain (nDCG): Assess ranking usefulness based on human-judged relevance.
Language Alignment Score: Quantify the semantic similarity between retrieved furniture metadata and the natural language description.
6. Conclusion
This challenge aims to advance research in multi-modal retrieval, spatial reasoning, and semantic understanding within the domain of intelligent 3D interior design. By integrating 3D spatial data and natural language inputs, we seek to push the boundaries of context-aware retrieval systems, facilitating practical applications in architecture, virtual staging, and smart home design.