I have a growing collection of high-resolution photos of farmed fish, each labeled according to the specific disease or healthy state shown. I need a robust image recognition model trained to spot those diseases quickly and accurately so the system can be deployed in a real-time monitoring pipeline for aquaculture. The job revolves around: • Cleaning and augmenting the dataset (currently JPG and PNG files, ~20 GB total). • Selecting or designing the most suitable CNN/transformer architecture in TensorFlow or PyTorch. • Training and validating the model, then fine-tuning until it meets or exceeds 92 % F1 on my held-out test set. • Packaging the finished model plus reproducible training scripts and clear inference instructions (Docker image or notebook is fine). Acceptance criteria 1. Trained model file(s) ready for GPU or CPU inference. 2. Python code that re-creates the final weights from raw images. 3. Evaluation report showing precision, recall and confusion matrix for each disease class. 4. Brief hand-off call or document to walk me through usage and future retraining steps. If you have prior work detecting objects or medical conditions in images—and can reference projects that achieved strong accuracy—let’s talk.