* Tuned an identification model on top of Fast R-CNN and YOLOv4 using augmented data and hyperparameter tuning. Utilized OpenCV and PyTorch on an internal linux server with multiple GPUs* Investigated and compared the Transformer architecture for potential application in current task. Presented findings and recommendations to team members for consideration in future projects* Accomplished 15th out of 70 position in the ICDAR 2021 document layout recognition team ranking, elevated overall accuracy from the baseline (85%) to 94%* Developed a strategy to leverage training data for identifying the best performing models for each box category, integrating results for the final output