How ashishpatel26/skin-lesions-detection-deep-learning works
Project type and purpose
• This is an open-source machine-learning research/experimentation project (not a production web app).
• Its goal is to classify skin lesions from dermatoscopic images using deep convolutional networks (baseline CNN, fine-tuned pretrained models, and an ensemble).
File structure
• Top-level:
• LICENSE
• README.md
• Project_Report.pdf
• images/ (dataset preview screenshots)
• Notebooks/ (each Notebook encapsulates one stage or experiment)
• Notebooks sub-folders:
• Skin_Cancer_EDA.ipynb: data ingestion & exploratory analysis
• Baseline_CNN.ipynb: training a simple CNN from scratch
• Fine_Tuning_*.ipynb: fine-tuning various pretrained architectures (VGG16, InceptionV3, InceptionResNet, DenseNet)
• Retraining_*.ipynb: full re-training on ImageNet weights
• Ensemble_Models.ipynb: combining best models into an ensemble
README for additional insights
• Describes the HAM10000 dataset as the source of images.
• Lists each notebook’s purpose in the pipeline.
• Mentions Keras 2.2.4/TensorFlow 1.11 and a known BatchNorm caveat.
• Presents final validation/test accuracies for each model variant.