EDISS is pleased to announce that a research paper co-authored by EDISS students has been accepted for publication at the IEEE Conference on Artificial Intelligence (IEEE CAI), a leading international venue for cutting-edge research in artificial intelligence.
The paper is co-authored by Ahmad Alkhaldi, Igli Balla, Rahma El Bouazzaoui, and Tirthendu Prosad Chakravorty, all EDISS doctoral students. The work was carried out under the co-supervision of Tristan Klempka (AILiveSim Oy), Hergys Rexha and Sébastien Lafond (Åbo Akademi University), highlighting the strong collaboration between academia and industry fostered within the EDISS programme.
The publication is titled Hybrid 3D Asset Retrieval via Contrastive Vision-Language Matching and Structured Prompt Parsing
Advancing 3D Asset Retrieval with Hybrid AI Methods
The accepted publication addresses a key challenge in simulation and virtual environment development: retrieving relevant 3D assets using natural language queries. Existing approaches often struggle to balance semantic flexibility with precise attribute control. Embedding-based methods typically fail when queries include multiple constraints, while metadata-based systems lack the ability to handle abstract or semantic descriptions.
To overcome these limitations, the authors propose a hybrid 3D asset retrieval framework that combines the strengths of both approaches. The system integrates vision–language models—such as CLIP, BLIP, and OpenCLIP—for semantic alignment, together with structured natural language parsing. Its dual-path architecture enables:
- Semantic matching through contrastive embeddings, and
- Explicit attribute filtering via a fine-tuned BERT model that extracts attribute–value pairs from user queries.
The framework was evaluated on a custom dataset of 500 multi-view 3D assets with structured annotations. Experimental results demonstrate that the hybrid approach significantly outperforms standalone retrieval baselines, achieving precision improvements of up to 357.7% for complex, multi-attribute queries.
Impact on Simulation and Automated Environment Creation
The proposed method enhances 3D asset retrieval workflows in simulation environments where both accuracy and fine-grained control are essential. By enabling more reliable and expressive natural language queries, the framework paves the way for automated environment creation and more efficient simulation pipelines.
This achievement reflects the EDISS programme’s commitment to high-impact research, interdisciplinary collaboration, and strong engagement with industry partners.
Congratulations to the authors on this outstanding accomplishment!
Last updated on 8 February 2026
