GabrieleGiudici/E-BARD-JerseyNumberRecognition

Title: Basketball Jersey Number Recognition (JNR) Dataset (Aggregated Base & Pose-Enhanced)

Relation to Paper: Developed as a key contribution of the E-BARD paper to support tasks like player tracking and identification. This repository aggregates both the standard JNR dataset and the pose-enhanced version designed to improve number readability through upper-body cropping.

Dataset Statistics:

• Base Dataset: * Contains 3,633 manually labelled frame stubs of detected players tracked with ByteTrack.
  • Each stub is a sequence of 16 consecutive frames.
  • Covers 57 distinct jersey numbers out of 101 possible classes, with 300 stubs annotated as "NaN" where the number was not visible.
• Pose-Enhanced Dataset: * Generated using MediaPipe to extract upper-body anatomical landmarks (shoulders and hips) for tighter torso cropping.
  • Strictly filtered to 983 stubs where all 16 frames were fully preserved to guarantee temporal context and legibility.