Object detector architecture research: compares YOLOv8m and RetinaNet+MiT-b0 on a custom-annotated ecology dataset with hard-negative mining and explainable AI galleries. Ablations include optimiser family, learning rate scheduling, progressive unfreezing and augmentations.
computer-vision deep-learning image-annotation tensorflow grad-cam object-detection ecology hyperparameter-tuning interpretability retinanet hard-negative-mining transfer-lea keras-tuner ablation-study yolov8 vision-trans eigencam
-
Updated
May 22, 2026 - Jupyter Notebook