Search papers, labs, and topics across Lattice.
The DS@GT team participated in the ImageCLEFmedical Caption 2026 challenge, focusing on two tasks: Concept Detection and Caption Prediction. For Concept Detection, they achieved first place with a late-fusion ensemble of ConvNeXt-V2, BiomedCLIP, and DenseNet-169, utilizing a novel "Honest Threshold Tuning" method to mitigate overfitting on rare concepts, resulting in a primary F1 score of 0.5790. In Caption Prediction, they explored various model configurations, including a fine-tuned Gemma-3 model that secured third place, demonstrating the effectiveness of diverse architectural strategies in medical image analysis.
A late-fusion ensemble approach not only topped the Concept Detection leaderboard but also highlighted the potential of training-free methods to match fine-tuned performance at a fraction of the cost.
We describe the DS@GT submissions to the ImageCLEFmedical Caption 2026 challenge, which continues a long-running benchmark on the ROCOv2 dataset with two tracks: Concept Detection (Task 1), assigning UMLS Concept Unique Identifiers (CUIs) to radiology images, and Caption Prediction (Task 2), generating natural-language captions. For Task 1, our primary submission was a three-way late-fusion ensemble of ConvNeXt-V2, BiomedCLIP ViT-B/16, and DenseNet-169 with a regularized''Honest Threshold Tuning''procedure designed to avoid validation overfitting on rare concepts; this submission ranked first on the official submission with a primary $F_1$ of $0.5790$ and a secondary $F_1$ of $0.9657$. In parallel, we submitted a training-free KNN retrieval pipeline over frozen BiomedCLIP embeddings, which reached a primary $F_1$ of $0.5780$ and a secondary $F_1$ of $0.9599$-essentially matching the fine-tuned ensemble on the primary track at a fraction of the cost. For Task 2, our submissions included a fine-tuned Gemma-3 27B model (overall $0.3571$, ranking third in the official submission), a fully fine-tuned BLIP pipeline with custom Vizwins merging ($0.3564$), and a zero-shot MedGemma-4B run with a PubMed-style prompt ($0.3186$), spanning a wide range of model scales and training costs. Code: https://github.com/dsgt-arc/imageclef-caption-2026.
