Pothole Detection UI

Run Detection

Pick Kaggle sample Or upload image Detection threshold

Recommended for the current model: 0.9029

Latest Result

No image has been analyzed in the UI yet.

Training Report

{
    "dataset_mode": "synthetic_plus_seed_positives_plus_unlabeled_road_negatives",
    "record_count": 32,
    "sample_count": 1413,
    "hard_negative_count": 63,
    "train_count": 1130,
    "validation_count": 283,
    "epochs": 70,
    "hidden_dim": 96,
    "patch_size": 32,
    "train_accuracy": 0.93893805309734512665187367019825614988803863525390625,
    "validation_accuracy": 0.932862190812720815102920823846943676471710205078125,
    "recommended_threshold": 0.9029000000000000358824081558850593864917755126953125,
    "seed_image_scores": {
        "download.jpg": 0.9981999999999999761968183520366437733173370361328125,
        "download (1).jpg": 0.9229000000000000536459765498875640332698822021484375
    },
    "demo_detection_count": 0,
    "history_tail": [
        {
            "epoch": 20,
            "loss": 0.304220546135860192560329551270115189254283905029296875,
            "accuracy": 0.94159292035398234332888023345731198787689208984375,
            "val_accuracy": 0.80212014134275622101455383017309941351413726806640625
        },
        {
            "epoch": 21,
            "loss": 0.30522735358866970312163857670384459197521209716796875,
            "accuracy": 0.93893805309734512665187367019825614988803863525390625,
            "val_accuracy": 0.932862190812720815102920823846943676471710205078125
        },
        {
            "epoch": 22,
            "loss": 0.294425165521360054921018445384106598794460296630859375,
            "accuracy": 0.94867256637168140276372696462203748524188995361328125,
            "val_accuracy": 0.932862190812720815102920823846943676471710205078125
        },
        {
            "epoch": 23,
            "loss": 0.289894838301481383435742600340745411813259124755859375,
            "accuracy": 0.951327433628318619440733527881093323230743408203125,
            "val_accuracy": 0.932862190812720815102920823846943676471710205078125
        },
        {
            "epoch": 24,
            "loss": 0.285700255474158115731597717967815697193145751953125,
            "accuracy": 0.954867256637168093647005662205629050731658935546875,
            "val_accuracy": 0.91519434628975260270777880577952601015567779541015625
        }
    ],
    "model_path": "C:\\xampp\\htdocs\\pothole_modle\\artifacts\\pothole_mlp.json",
    "demo_scene": "C:\\xampp\\htdocs\\pothole_modle\\artifacts\\demo_scene.png",
    "demo_prediction": "C:\\xampp\\htdocs\\pothole_modle\\artifacts\\demo_prediction.png",
    "kaggle_notebook": "https://www.kaggle.com/code/arnavaku/pothole-detection-01-18-2026/notebook",
    "kaggle_input": "https://www.kaggle.com/code/arnavaku/pothole-detection-01-18-2026/input",
    "note": "The trainer used YOLO labels if present. Otherwise it falls back to synthetic data and can also mine unlabeled road images as hard negatives."
}