Skip to main content

Crate yscv_eval

Crate yscv_eval 

Source
Expand description

§yscv-eval

Evaluation metrics for classification, detection, and tracking. Dataset adapters for COCO, Pascal VOC, and CSV.

ⓘ
use yscv_eval::*;

let ap = average_precision(&predictions, &ground_truths, 0.5);
let report = classification_report(&predicted_labels, &true_labels);
println!("{}", report);

§Metrics

37 public metric/eval functions across the crate:

TaskMetrics
Classificationaccuracy, precision, recall, F1, confusion matrix, Cohen’s kappa, ROC curve + AUC, average_precision, top_k_accuracy, classification_report
DetectionmAP, COCO mAP, AP@IoU, precision-recall, evaluate_detections{,_coco,_from_dataset}
TrackingMOTA, MOTP, HOTA, IDF1, evaluate_tracking{,_from_dataset}
RegressionMAE, RMSE, MAPE, R²
Image quality / segmentationPSNR, SSIM, dice_score, mean_iou, per_class_iou
Counting / pipeline / cameracounting metrics, pipeline benchmark thresholds, camera diagnostics validation

§Dataset Adapters (8)

Under src/dataset/:

  • COCO — parse_detection_dataset_coco, load_detection_dataset_coco_files
  • JSONL — detection + tracking
  • MOT — parse_tracking_dataset_mot, load_tracking_dataset_mot_txt_files
  • OpenImages — CSV pair (parse_detection_dataset_openimages_csv)
  • VOC — XML directories
  • YOLO — label directories
  • KITTI — label directories
  • WIDERFACE — TXT pair

§Tests

95 tests covering metric correctness, edge cases, dataset parsing.

Structs§

BenchmarkViolation
CameraDiagnosticsCapture
CameraDiagnosticsDevice
CameraDiagnosticsFirstFrame
CameraDiagnosticsReport
CameraDiagnosticsRequested
CameraDiagnosticsThresholds
CameraDiagnosticsTiming
CameraDiagnosticsViolation
CocoMetrics
CountingMetrics
DetectionDatasetFrame
DetectionEvalConfig
DetectionFrame
DetectionMetrics
GroundTruthTrack
LabeledBox
PipelineBenchmarkReport
PipelineBenchmarkThresholds
PipelineDurations
StageThresholds
TimingStats
TrackingDatasetFrame
TrackingEvalConfig
TrackingFrame
TrackingMetrics

Enums§

EvalError
F1Average
Averaging strategy for multi-class F1 score.

Constants§

CRATE_ID

Functions§

accuracy
Compute classification accuracy as the fraction of correct predictions.
auc
Area under the curve using the trapezoidal rule.
average_precision
Compute average precision (area under the precision-recall curve) using the trapezoidal rule.
classification_report
Generate a human-readable classification report (similar to scikit-learn’s classification_report).
cohens_kappa
Cohen’s kappa coefficient measuring inter-annotator agreement.
confusion_matrix
Compute a confusion matrix for num_classes classes.
detection_frames_as_view
dice_score
Per-class Dice coefficient: 2 * |pred ∩ target| / (|pred| + |target|).
evaluate_counts
evaluate_detections
evaluate_detections_coco
Evaluate detections using COCO-style multi-threshold metrics.
evaluate_detections_from_dataset
evaluate_tracking
evaluate_tracking_from_dataset
f1_score
Compute F1 score with the specified averaging strategy.
hota
Higher Order Tracking Accuracy.
idf1
Identity F1 score: measures how well predicted IDs match GT IDs across frames.
load_camera_diagnostics_report_json_file
load_detection_dataset_coco_files
load_detection_dataset_jsonl_file
load_detection_dataset_kitti_label_dirs
load_detection_dataset_openimages_csv_files
load_detection_dataset_voc_xml_dirs
load_detection_dataset_widerface_files
load_detection_dataset_yolo_label_dirs
load_tracking_dataset_jsonl_file
load_tracking_dataset_mot_txt_files
mae
Mean absolute error.
mape
Mean absolute percentage error, skipping pairs where the target is zero.
mean_iou
Mean Intersection over Union for semantic segmentation.
parse_camera_diagnostics_report_json
parse_detection_dataset_coco
parse_detection_dataset_jsonl
parse_detection_dataset_openimages_csv
parse_detection_dataset_widerface
parse_pipeline_benchmark_thresholds
parse_tracking_dataset_jsonl
parse_tracking_dataset_mot
per_class_iou
Per-class Intersection over Union: |pred ∩ target| / |pred ∪ target|.
per_class_precision_recall
Compute per-class precision and recall from a confusion matrix.
precision_recall_curve
Compute precision-recall curve from binary classification scores and labels.
psnr
Peak Signal-to-Noise Ratio between two images.
r2_score
Coefficient of determination: 1 - SS_res / SS_tot.
rmse
Root mean squared error.
roc_curve
Compute ROC curve from binary classification scores and labels.
ssim
Structural Similarity Index (SSIM) between two grayscale images.
summarize_durations
summarize_pipeline_durations
top_k_accuracy
Top-k accuracy: fraction of samples where the correct label is in the top-k predictions.
tracking_frames_as_view
validate_camera_diagnostics_report
validate_pipeline_benchmark_thresholds