How yadavdevam481-ops/dermassit-gen works
DermAssist-GEN is a research workflow for preparing HAM10000 data, training and evaluating a seven-class skin-lesion image classifier, and producing/evaluating synthetic dermoscopy images with conditional GAN or Stable Diffusion LoRA models. The demo exposes classification and generation guidance; it does not implement the README’s claimed treatment prescription, and its generation tab does not generate images itself. Paths follow the supplied flat tree; model and dataset internals are represented only where sampled source supports them.
Main entry points: train_classifier.py Classifier training · evaluate_classifier.py Classifier evaluation · classifier.py EfficientNet classifier · prepare_ham10000.py Dataset preparation
Components 20
HAM10000 data
- Dataset preparation
prepare_ham10000.py - Split image index
- Dermoscopy dataset
dataset.py
Image classification
- EfficientNet classifier
classifier.py - Training losses
losses.py - Classifier training
train_classifier.py - Research demo
app.py
Image synthesis
- Conditional GAN training
train_gan.py - Conditional GAN
cgan.py - GAN image generation
generate_gan.py - LoRA training
train_diffusion_lora.py - Diffusion generation
generate_diffusion.py - Synthetic images
Research evaluation
- Classifier evaluation
evaluate_classifier.py - Classification metrics
metrics.py - Generation evaluation
evaluate_generation.py
Other components
- Researcher
- HAM10000 files
- Stable Diffusion model
- Real image set
Connections 28
- Researcher → Dataset preparation: provides metadata
- HAM10000 files → Dataset preparation: provides images
- Dataset preparation → Split image index: writes splits
- Split image index → Dermoscopy dataset: indexes images
- Dermoscopy dataset → Classifier training: loads batches
- Classifier training → EfficientNet classifier: trains
- Classifier training → Training losses: uses
- Classifier training → Classification metrics: scores validation
- Researcher → Research demo: uploads image
- Research demo → EfficientNet classifier: classifies
- EfficientNet classifier → Research demo: returns probabilities
- Split image index → Conditional GAN training: indexes training images
- Conditional GAN training → Conditional GAN: trains
- GAN image generation → Conditional GAN: loads checkpoint
- GAN image generation → Synthetic images: writes images
- Split image index → LoRA training: indexes training images
- HAM10000 files → LoRA training: supplies image files
- Stable Diffusion model → LoRA training: provides pretrained weights
- Diffusion generation → Stable Diffusion model: loads pretrained pipeline
- Diffusion generation → Synthetic images: writes images
- Synthetic images → Dermoscopy dataset: augments training data
- Researcher → Classifier evaluation: requests evaluation
- Split image index → Classifier evaluation: indexes evaluation images
- Classifier evaluation → EfficientNet classifier: loads checkpoint
- Classifier evaluation → Classification metrics: computes metrics
- Researcher → Generation evaluation: requests evaluation
- Real image set → Generation evaluation: provides reference images
- Synthetic images → Generation evaluation: provides generated images