How alok-senapati/cifar100-cnn works
CIFAR-100 CNN is a PyTorch image-classification project whose main workflow configures a run, loads and normalizes deterministic train/validation/test data, trains one of three CNN variants, selects the lowest-validation-loss checkpoint, and evaluates it. A Streamlit dashboard separately discovers saved runs and serves test-image predictions. The README documents augmentation, schedulers, BatchNorm, metrics, and checkpoint artifacts. Some script and trainer details are unsampled; edges use README statements or visible/import-index evidence rather than assuming internal calls.
Main entry points: trainer.py Training and evaluation · train_gap_classifier.py GAP run · train_baseline.py Baseline run · train_reduced_pooling.py Reduced-pooling run
Components 17
Training workflow
- Run configuration
src/cifar100_cnn/args/parser_classes.py - Baseline run
scripts/train_baseline.py - GAP run
scripts/train_gap_classifier.py - Reduced-pooling run
scripts/train_reduced_pooling.py
Dataset preparation
- CIFAR-100 data
src/cifar100_cnn/data/loader.py - Image transforms
src/cifar100_cnn/data/transforms.py
CNN architectures
- Baseline CNN
src/cifar100_cnn/model/baseline.py - GAP CNN
src/cifar100_cnn/model/gap_classifier.py - Reduced-pooling CNN
src/cifar100_cnn/model/reduced_pooling.py
Training and evaluation
- Training and evaluation
src/cifar100_cnn/model/trainer.py - Checkpoint reconstruction
src/cifar100_cnn/model/loader.py - Gradient diagnostics
src/cifar100_cnn/utils/diagnose.py - Console formatting
src/cifar100_cnn/utils/printer.py - Epoch timing
src/cifar100_cnn/utils/timer.py - Metric and result plots
src/cifar100_cnn/utils/visualizer.py
Inference dashboard
- Streamlit inference
app.py
Other components
- Researcher
Connections 23
- Researcher → Baseline run: starts training
README.md - Researcher → GAP run: starts training
README.md - Researcher → Reduced-pooling run: starts training
README.md - Researcher → Streamlit inference: opens dashboard
README.md - Baseline run → Run configuration: reads options
scripts/train_baseline.py - GAP run → Run configuration: reads options
scripts/train_gap_classifier.py - Reduced-pooling run → Run configuration: reads options
scripts/train_reduced_pooling.py - Baseline run → CIFAR-100 data: loads splits
scripts/train_baseline.py - GAP run → CIFAR-100 data: loads splits
scripts/train_gap_classifier.py - Reduced-pooling run → CIFAR-100 data: loads splits
scripts/train_reduced_pooling.py - GAP run → Image transforms: configures augmentation
scripts/train_gap_classifier.py - Baseline run → Baseline CNN: trains variant
README.md - GAP run → GAP CNN: trains variant
README.md - Reduced-pooling run → Reduced-pooling CNN: trains variant
README.md - Baseline run → Training and evaluation: invokes training
scripts/train_baseline.py - GAP run → Training and evaluation: invokes training
scripts/train_gap_classifier.py - Reduced-pooling run → Training and evaluation: invokes training
scripts/train_reduced_pooling.py - Training and evaluation → Epoch timing: measures epochs
src/cifar100_cnn/model/trainer.py - Training and evaluation → Gradient diagnostics: uses diagnostics
src/cifar100_cnn/model/trainer.py - Training and evaluation → Console formatting: formats output
src/cifar100_cnn/model/trainer.py - Training and evaluation → Metric and result plots: renders metrics
src/cifar100_cnn/model/trainer.py - Streamlit inference → Checkpoint reconstruction: loads selected model
app.py - Streamlit inference → CIFAR-100 data: loads test data
app.py