jeremymboe/datasetandoutput_fpPCD_Xray

Chest X-Ray Restoration & Classification — Dataset + Pipeline Outputs

Companion dataset for the project Restorasi dan Klasifikasi Citra X-Ray Dada Berbasis Filter Spasial, Frekuensi, dan Peningkatan ROI (GitHub).

Contains the raw input dataset (data/) and every artifact generated by the pipeline (output/): filtered/enhanced images, classification logs, and evaluation reports.

Contents

data/
├── corrupted/              2500 degraded chest X-ray PNGs (pipeline input)
├── groundtruth/            2500 clean reference PNGs, same filenames as corrupted/
└── balanced_2500.csv       labels ("Image Index", "Finding Labels", ...)

output/
├── 01_preprocessing/       resized + normalized corrupted & groundtruth images
├── 02_spatial_filtering/   Gaussian LPF, Unsharp Masking + best_params.json
├── 03_frequency_filtering/ FFT spectrum, Butterworth LPF/HPF, Inverse FFT + best_params.json
├── 04_roi_enhancement/     CLAHE, morphology, histogram equalization, gamma correction
├── 05_classification/      per-scenario training logs (checkpoints excluded — too large)
└── 06_evaluation/          PSNR/SSIM metrics, classification metrics, summary_report.txt

Model checkpoints (.pth) are not included in this upload to keep the repo lean — retrain locally via the pipeline if you need them.

Source

2500 chest X-ray images, NIH ChestX-ray14 subset, labeled with one finding: Atelectasis, Effusion, Infiltration, No Finding, Nodule, Pneumothorax.

groundtruth/ is the clean reference used to score PSNR/SSIM and select filter parameters; corrupted/ is the actual pipeline input.

Usage

hf download jeremymboe/datasetandoutput_fpPCD_Xray --repo-type dataset --local-dir .

Full pipeline code, methodology, and results table: see the GitHub repo README.