InceptionV3 Food-101 Image Classification
Summary I fine-tuned an InceptionV3 model on the Food-101 dataset with mixed-precision training and a custom learning-rate schedule, using checkpointing to stabilize improvements across epochs on my RTX 3090. Target was robust top-1 accuracy with strong generalization to unseen food categories.
Highlights
Mixed-precision training to accelerate throughput and reduce VRAM use
Cosine LR schedule with warmup and checkpointing for stable convergence
Reproducible training scripts and evaluation utilities
Stack Keras, TensorFlow, Python, RTX 3090, CUDA, cuDNN
Links
Repo: https://github.com/jengalbert2022/FoodNutritionEstimator
Project link: https://github.com/jengalbert2022/FoodNutritionEstimator