I have a collection of raw image data that needs to be turned into a production-ready image classification model. The project is strictly data-analysis-oriented: everything from exploring the dataset through to delivering a trained network and clear performance metrics. Scope of work • Inspect and prepare the images (deduplication, resizing, augmentation where it actually improves results). • Select or design a deep-learning architecture that suits the dataset size and class balance—transfer learning with TensorFlow or PyTorch is perfectly fine if it speeds up convergence. • Train, validate and test the model, then hand over the full training notebook or script with comments so I can reproduce results on my own machine. • Provide a concise evaluation report: final accuracy, confusion matrix, and any relevant error analysis that explains common misclassifications. Acceptance criteria • Minimum overall accuracy of 90 % on a held-out test set I will provide. • All code must run end-to-end on a standard GPU environment using Python (3.8+). • Final deliverables include the trained weights, environment file/requirements.txt, and the well-documented source code or notebook. If any assumptions about the data or labels are needed, let me know early so we can pin them down before you start training.