How sakiyo212/taresnet works
TAresnet is a diabetic-retinopathy image-classification research project: its visible runtime script loads labeled images, trains a pretrained ResNet-152 classifier, validates it, and saves the best checkpoint. The README also describes Grad-CAM analysis, but no implementation for that capability appears in the tree. The listed prediction CSVs and research documents are present, but their generation or use is not established by the sampled source, so they are not wired into the runtime flow.
Main entry points: main.py Training script
Components 14
Training workflow
- Training script
main.py - Dataset loading
main.py - Image preprocessing
main.py - Balanced split
main.py - ResNet-152 model
main.py - Model training
main.py - Model validation
main.py
Research artifacts
- Best checkpoint
main.py - Default predictions
default prediction/resnet152val_sorted.csv - Weighted predictions
class_weighted pred/resnet152valClass_sorted.csv
Model analysis
- Grad-CAM analysis
README.md
Other components
- Researcher
- Labeled image data
- Torch Hub weights
Connections 13
- Researcher → Training script: runs
main.py - Training script → Dataset loading: loads data
main.py - Labeled image data → Dataset loading: provides labels/images
main.py - Dataset loading → Image preprocessing: transforms images
main.py - Dataset loading → Balanced split: splits labels
main.py - Image preprocessing → Model training: feeds batches
main.py - Balanced split → Model training: feeds training set
main.py - Balanced split → Model validation: feeds validation set
main.py - Training script → ResNet-152 model: creates
main.py - ResNet-152 model → Torch Hub weights: loads weights
main.py - Model training → ResNet-152 model: trains
main.py - Model validation → ResNet-152 model: evaluates
main.py - Training script → Best checkpoint: saves best
main.py