---
license: mit
tags:
- computer-vision
- sports-analytics
- jersey-recognition
- temporal-modeling
- lstm
- bilstm
- pytorch
datasets:
- custom
metrics:
- accuracy
model-index:
- name: jersey-number-recognition
  results:
  - task:
      type: image-classification
      name: Jersey Number Recognition
    metrics:
    - type: accuracy
      value: 92.12
      name: Full Number Accuracy
    - type: accuracy
      value: 98.63
      name: Tens Digit Accuracy
    - type: accuracy
      value: 93.04
      name: Units Digit Accuracy
---

# Jersey Number Recognition - Temporal BiLSTM Model

<div align="center">
  <img src="https://img.shields.io/badge/Accuracy-92.12%25-success" alt="Accuracy"/>
  <img src="https://img.shields.io/badge/PyTorch-2.0+-red" alt="PyTorch"/>
  <img src="https://img.shields.io/badge/License-MIT-blue" alt="License"/>
</div>

## Model Description

A BiLSTM-based temporal model for recognizing jersey numbers from video sequences, achieving **92.12% accuracy** - a **43% improvement** over single-frame baselines.

### Key Features

- 🎯 **92.12%** full number accuracy
- 🎯 **98.63%** tens digit accuracy  
- 🎯 **93.04%** units digit accuracy
- 🎯 **89%** temporal stability across player tracks
- 🎯 Compositional generalization to 100 classes (00-99)

## Model Architecture
```
Input Sequence [8 × 3 × 128 × 128]
    ↓
EfficientNet-B0 Backbone (shared weights)
    ↓
256-D Embeddings [8 × 256]
    ↓
2-Layer Bidirectional LSTM (hidden: 128)
    ↓
Concatenated Hidden States [512]
    ↓
    ├─→ Tens Digit Head (10 classes)
    └─→ Units Digit Head (10 classes)
```

**Parameters**: 5.1M  
**Model Size**: 20.3 MB

## Intended Use

### Primary Use Cases

- Jersey number recognition in sports analytics
- Temporal sequence modeling for visual recognition
- Research in compositional generalization

### Out-of-Scope Uses

- Real-time applications (not optimized for inference speed)
- Non-sports contexts without fine-tuning
- Privacy-sensitive applications

## How to Use

### Installation
```bash
pip install torch torchvision pillow
```

### Quick Start
```python
import torch
from PIL import Image
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(
    repo_id="prxkc/jersey-number-recognition",
    filename="best_temporal.pt"
)

# Load checkpoint
checkpoint = torch.load(model_path, map_location='cpu')

# Note: You'll need the model architecture code
# See GitHub repository for complete implementation
# GitHub: https://github.com/prxkc/jersey-number-recognition
```

### Complete Example

For complete usage with model architecture, see the [GitHub Repository](https://github.com/prxkc/jersey-number-recognition).

## Training Data

- **Dataset**: Custom jersey number dataset (subset)
- **Train samples**: 4,096 sequences
- **Validation samples**: 860 sequences
- **Test samples**: 876 sequences
- **Classes**: 10 jersey numbers (subset of 00-99)

### Data Preprocessing

- Frames resized to 128×128 pixels
- Pad-to-square transformation
- ImageNet normalization
- 8 frames uniformly sampled per sequence

## Training Procedure

### Hyperparameters

- **Backbone**: EfficientNet-B0 (pretrained)
- **Optimizer**: AdamW (lr=2e-4, weight_decay=1e-3)
- **Scheduler**: Cosine annealing
- **Batch size**: 32 (temporal), 128 (anchor)
- **Epochs**: 10 (temporal), 4 (anchor warmstart)
- **Mixed precision**: Enabled (AMP)

### Training Strategy

1. **Warmstart**: Train anchor-only baseline (4 epochs)
2. **Temporal training**: BiLSTM model (10 epochs)
3. **Backbone freezing**: First 2 epochs
4. **Balanced sampling**: Digit-level balancing

## Evaluation Results

### Test Set Performance

| Metric | Anchor (Baseline) | Temporal (Ours) | Improvement |
|--------|-------------------|-----------------|-------------|
| Full Number Acc | 48.97% | **92.12%** | +43.15% |
| Tens Digit Acc | 92.81% | **98.63%** | +5.82% |
| Units Digit Acc | 53.31% | **93.04%** | +39.73% |
| Loss | 1.358 | **0.336** | -75.3% |

### Temporal Stability

- **89%** of tracks had zero prediction flips
- **Average 0.11 flips** per track
- Significant improvement over single-frame predictions

### Per-Class Results

| Jersey # | Test Sequences | Accuracy |
|----------|----------------|----------|
| 4 | 164 | 95.73% |
| 6 | 134 | 94.78% |
| 8 | 301 | 90.70% |
| 9 | 216 | 90.28% |
| 48 | 4 | 100.00% |
| 49 | 19 | 89.47% |
| 66 | 19 | 100.00% |
| 89 | 16 | 93.75% |

## Limitations

- Trained on limited jersey number subset (10 classes)
- Not optimized for real-time inference
- Requires 8-frame sequences (not single images)
- Performance may degrade on very different visual conditions

## Contact

- **Author**: Shakil Islam Shanto
- **GitHub**: [@prxkc](https://github.com/prxkc)