CNN Image Classification — Custom Convolutional Network Build

I will build and train a convolutional neural network for your image classification task — covering data augmentation pipeline, CNN architecture design or backbone selection, class imbalance handling, training with validation monitoring, evaluation with confusion matrix and per-class metrics, and a deployment-ready model with an inference function. Image classification CNNs built without data augmentation systematically overfit to the training set; without per-class evaluation metrics they hide the fact that the model performs well on majority classes and poorly on the minority classes most likely to be your actual business priority.

The build covers image loading and augmentation pipeline (random crop, flip, colour jitter, normalise), CNN architecture or transfer learning backbone, class weight handling for imbalanced datasets, training with validation monitoring, per-class precision/recall/F1 evaluation, confusion matrix, and an inference function accepting a single image and returning the top-k predictions.