MARSAIL
(Motor AI Recognition Solution AI Lab)
MARS is a state-of-the-art deep learning model for car damage instance segmentation. By leveraging sequential quadtree attention, it surpasses existing methods (Mask R-CNN, PointRend, Mask Transfiner) with notable maskAP gains.
Presented at ICIAP 2023, Udine, Italy ๐ฎ๐น
MARSAIL (Motor AI Recognition Solution Artificial Intelligence Laboratory) is dedicated to pioneering research at the intersection of computer vision, transformers, and automotive AI.
Our mission is to revolutionize the automotive insurance and repair industries through AI-driven automation, delivering breakthroughs in segmentation, localization, and decision intelligence.
MARS delivers superior segmentation accuracy:
At the forefront of automotive AI innovation, all pioneering research at MARSAIL is led by Dr. Kao (Teerapong Panboonyuen), our visionary Head of AI Laboratory. Under his guidance, these groundbreaking works have advanced the fields of car damage evaluation, segmentation, and document understanding in the automotive domain.
A pioneering transformer-based architecture designed to act as a teacher model for vehicle damage evaluation tasks. ALBERT provides rich contextual embeddings and accurate localization signals that guide downstream student models. [Read on arXiv]
@article{panboonyuen2025albert, title={ALBERT: Advanced Localization and Bidirectional Encoder Representations from Transformers for Automotive Damage Evaluation}, author={Panboonyuen, Teerapong}, journal={arXiv preprint arXiv:2506.10524}, year={2025} }
Trained under the supervision of ALBERT, SLICK leverages knowledge distillation and selective localization to excel in fine-grained car damage segmentation tasks โ a student model optimized for real-world insurance workflows. [Read on arXiv]
@article{panboonyuen2025slick, title={SLICK: Selective Localization and Instance Calibration for Knowledge-Enhanced Car Damage Segmentation in Automotive Insurance}, author={Panboonyuen, Teerapong}, journal={arXiv preprint arXiv:2506.10528}, year={2025} }
Combines deformable attention mechanisms and Retrieval-Augmented Generation (RAG) for robust end-to-end document understanding in the automotive industry. [Read on arXiv]
@inproceedings{marsail2025dota, title={DOTA: Deformable Optimized Transformer Architecture for End-to-End Text Recognition with Retrieval-Augmented Generation}, author={Panboonyuen, Teerapong as co-author and lead researcher}, booktitle={2025 17th International Conference on Knowledge and Smart Technology (KST)}, pages={301--306}, year={2025}, organization={IEEE} }
Introduces a sequential quadtree attention strategy to refine segmentation masks, enhancing instance-level car damage detection accuracy. [Read on Springer]
@inproceedings{panboonyuen2023mars, title={Mars: Mask attention refinement with sequential quadtree nodes for car damage instance segmentation}, author={Panboonyuen, Teerapong}, booktitle={International Conference on Image Analysis and Processing}, pages={28--38}, year={2023}, organization={Springer} }
git clone https://github.com/kaopanboonyuen/MARS.git cd MARS python3 -m venv mars-env source mars-env/bin/activate pip install -r requirements.txt
@inproceedings{panboonyuen2023mars, title={MARS: Mask Attention Refinement with Sequential Quadtree Nodes}, author={Panboonyuen, Teerapong}, booktitle={ICIAP}, year={2023}, publisher={Springer} }
Explore how our lab is shaping the future of automotive AI, blending deep vision models with smart insurance tech. Get the full story, insights, and breakthroughs from the MARSAIL team in our feature blog post.
Read the Full BlogWe are deeply grateful for the invaluable support of our sponsors, partners, and collaborators. Their investment and belief in our mission empower MARSAIL to push boundaries in automotive AI.