How brookww/papercnn works
PaperCNN classifies research fields from paper first-page images. Users either supply an image and load a checkpoint for a prediction, or provide dataset CSVs and images to check the data, train and validate a model, then evaluate a saved checkpoint on a split. The graph separates data preparation, the classifier, training, and prediction/evaluation. Source excerpts show the main calls; README statements support documented capabilities not fully visible in the excerpts.
Main entry points: train.py Training loop · predict.py Split evaluation · data.py Data loaders · model.py PaperCNN
Components 16
Data preparation
- Defaults
config.py - CSV validation
data.py - Train/val split
data.py - Image preprocessing
data.py - Data loaders
data.py - Data checks
check.py
Classifier
- PaperCNN
model.py - Batch update
model.py - Model checks
check.py
Training
- Training loop
train.py - Split metrics
train.py - Saved checkpoint
train.py
Prediction and evaluation
- Single-image prediction
predict.py - Split evaluation
predict.py
Other components
- User
- Dataset files
Connections 26
- User → Training loop: starts training
README.md - User → Single-image prediction: requests prediction
README.md - User → Split evaluation: requests evaluation
README.md - Dataset files → CSV validation: provides CSVs
data.py - Dataset files → Image preprocessing: provides images
data.py - CSV validation → Train/val split: supplies rows
data.py - Train/val split → Data loaders: builds datasets
data.py - Image preprocessing → Data loaders: loads samples
data.py - Defaults → Training loop: sets defaults
train.py - Defaults → Data loaders: sets image size
train.py - Data loaders → Training loop: supplies batches
train.py - Training loop → PaperCNN: initializes model
train.py - Training loop → Batch update: updates batches
train.py - Batch update → PaperCNN: runs forward pass
model.py - Training loop → Split metrics: evaluates validation
train.py - Split metrics → PaperCNN: runs inference
train.py - Training loop → Saved checkpoint: saves best model
train.py - Training loop → Model checks: runs checks
train.py - Training loop → Data checks: supports data checks
README.md - Single-image prediction → Saved checkpoint: loads model
predict.py - Single-image prediction → Image preprocessing: preprocesses image
predict.py - Single-image prediction → PaperCNN: classifies image
predict.py - Split evaluation → Saved checkpoint: loads model and split
predict.py - Split evaluation → CSV validation: reads test CSV
predict.py - Split evaluation → Data loaders: builds evaluation loader
predict.py - Split evaluation → Split metrics: computes metrics
predict.py