Back to projects
Fashion-MNIST Vision & Grad-CAM Analysis
A deep learning image classification project comparing a custom CNN with fine-tuned pretrained vision models, featuring Grad-CAM explainability.
PythonPyTorchComputer VisionGrad-CAMJupyter Notebook
Constructed and fine-tuned deep learning architectures on the FashionMNIST dataset to evaluate transfer learning against baseline computer vision models. Comparative benchmarks were established between a custom CNN built from scratch and a fine-tuned pretrained vision backbone.
Integrated Gradient-weighted Class Activation Mapping (Grad-CAM) to generate visual explainability maps. Analyzed feature extraction heatmaps to understand model attention, validate spatial decision boundaries, and document performance trade-offs in a comprehensive analytical report.