reskyz1/pad-ufes-20-experiments

How reskyz1/pad-ufes-20-experiments works

PAD-UFES-20 is a PyTorch research project for skin-lesion classification and controlled label-noise experiments. Its main workflow prepares patient-separated data, trains a cross-entropy or robust-method model, evaluates predictions per lesion, and can fit temperature scaling on validation predictions before test-set analysis. The graph also preserves noise-manifest generation and decision-curve analysis. Relationships are limited to those visible in excerpts, the source index, or the README; unsampled implementation details are not assumed.

Main entry points: train.py CE training · robust_train.py Robust training · evaluate.py Run evaluation · data.py Datasets and transforms

Components 18

Data preparation

Prepare data pad_ufes20_idn64/src/pad_idn/prepare.py
Patient splits pad_ufes20_idn64/src/pad_idn/splits.py
Datasets and transforms pad_ufes20_idn64/src/pad_idn/data.py
IDN generation pad_ufes20_idn64/src/pad_idn/noise.py
Noise manifests pad_ufes20_idn64/src/pad_idn/noise_manifest.py

Model experiments

CE training pad_ufes20_idn64/src/pad_idn/train.py
Robust training pad_ufes20_idn64/src/pad_idn/robust_train.py
Robust losses pad_ufes20_idn64/src/pad_idn/robust_methods.py
ResNet models pad_ufes20_idn64/src/pad_idn/models.py
Train and predict loops pad_ufes20_idn64/src/pad_idn/engine.py

Evaluation and calibration

Run evaluation pad_ufes20_idn64/src/pad_idn/evaluate.py
Lesion metrics pad_ufes20_idn64/src/pad_idn/metrics.py
Temperature scaling pad_ufes20_idn64/src/pad_idn/calibration.py
Decision curves pad_ufes20_idn64/src/pad_idn/decision_curve.py
Experiment results pad_ufes20_idn64/src/pad_idn/results.py

Shared runtime

Shared utilities pad_ufes20_idn64/src/pad_idn/utils.py

Other components

Researcher
PAD-UFES-20 data

Connections 22

Researcher → Prepare data: prepares data README.md
PAD-UFES-20 data → Prepare data: provides images
Researcher → CE training: runs training README.md
Prepare data → Patient splits: creates inputs
Patient splits → Datasets and transforms: defines partitions
IDN generation → ResNet models: extracts features pad_ufes20_idn64/src/pad_idn/noise.py
IDN generation → Noise manifests: uses manifests pad_ufes20_idn64/src/pad_idn/noise.py
CE training → Datasets and transforms: builds loaders pad_ufes20_idn64/src/pad_idn/train.py
CE training → ResNet models: builds model pad_ufes20_idn64/src/pad_idn/train.py
CE training → Train and predict loops: trains and predicts pad_ufes20_idn64/src/pad_idn/train.py
Robust training → Datasets and transforms: builds loaders pad_ufes20_idn64/src/pad_idn/robust_train.py
Robust training → ResNet models: builds models pad_ufes20_idn64/src/pad_idn/robust_train.py
Robust training → Robust losses: uses losses pad_ufes20_idn64/src/pad_idn/robust_train.py
Robust training → Train and predict loops: uses training loops pad_ufes20_idn64/src/pad_idn/robust_train.py
Run evaluation → Datasets and transforms: loads evaluation data pad_ufes20_idn64/src/pad_idn/evaluate.py
Run evaluation → ResNet models: builds model pad_ufes20_idn64/src/pad_idn/evaluate.py
Run evaluation → Train and predict loops: predicts images pad_ufes20_idn64/src/pad_idn/evaluate.py
Run evaluation → Lesion metrics: computes metrics pad_ufes20_idn64/src/pad_idn/evaluate.py
Researcher → Temperature scaling: calibrates predictions README.md
Temperature scaling → Lesion metrics: reports metrics pad_ufes20_idn64/src/pad_idn/calibration.py
Temperature scaling → Decision curves: computes curves pad_ufes20_idn64/src/pad_idn/calibration.py
CE training → Lesion metrics: evaluates predictions pad_ufes20_idn64/src/pad_idn/train.py

Drawn by GitDiagram on November 10, 2026 from the repository's file tree, README and source files. Read as Markdown ·