How tornadomeet/resnet works
This repository reproduces ResNet-v2 training and provides single-image classification with trained checkpoints. The main workflow is prepared CIFAR-10 or ImageNet record files into training and validation, build the selected network, train/evaluate and save checkpoints, then optionally classify an image. It also supports continued training, memory-saving integration, and plotting accuracy curves. Training, model construction, prediction, and plotting are visible in the sampled sources; dataset preparation and reported model results are documented in the README, not implemented in the shown files.
Main entry points: train_resnet.py Training runner · predict.py Image classifier · plot_curve.py Curve plotter · synset.txt Class labels
Components 13
Network
- ResNet-v2 symbol
symbol_resnet.py
Training
- Prepared records
- Training runner
train_resnet.py - Record iterators
train_resnet.py - Model checkpoints
train_resnet.py
Inference
- Input image
- Class labels
predict/synset.txt - Image classifier
predict/predict.py - Top-1 / top-5
Analysis
- Training logs
- Curve plotter
log/plot_curve.py
Other components
- Researcher
- MXNet runtime
Connections 16
- Researcher → Prepared records: prepares
README.md - Prepared records → Record iterators: provides records
train_resnet.py - Training runner → Record iterators: configures
train_resnet.py - Training runner → ResNet-v2 symbol: builds
train_resnet.py - Training runner → MXNet runtime: trains and evaluates
train_resnet.py - Training runner → Model checkpoints: saves / loads
train_resnet.py - Researcher → Training runner: starts training
README.md - Researcher → Input image: selects
- Researcher → Image classifier: starts prediction
predict/predict.py - Model checkpoints → Image classifier: loads parameters
predict/predict.py - Input image → Image classifier: classifies
predict/predict.py - Class labels → Image classifier: maps classes
predict/predict.py - Image classifier → Top-1 / top-5: prints rankings
predict/predict.py - Training runner → Training logs: produces training logs
README.md - Researcher → Curve plotter: requests plot
README.md - Training logs → Curve plotter: parses
log/plot_curve.py