Add YOLO11s FER2013 ensemble (2 weights + inference script + README)

159b2dd verified 7 months ago

917 Bytes

library_name: ultralytics
tags:
  - image-classification
  - facial-expression-recognition
  - fer2013
metrics:
  - accuracy

FER2013 Emotion Classification (YOLO11s Ensemble)

This repo contains an ensemble of two Ultralytics YOLO11s classification checkpoints. The final prediction is computed by averaging softmax probabilities from both models.

Test Results (FER2013 test set, 7,178 images)

Top-1 Accuracy (ensemble): 70.41%

Per-class accuracy (TEST, ensemble):

angry: 63.99%
disgust: 61.26%
fear: 49.51%
happy: 89.85%
neutral: 70.88%
sad: 57.82%
surprise: 81.47%

Files

weights/modelA_yolo11s_96.pt (YOLO11s trained @96)
weights/modelB_yolo11s_128.pt (YOLO11s fine-tuned @128)
ensemble_predict.py (simple CLI for ensemble inference)

Usage

pip install ultralytics numpy
python ensemble_predict.py --image path/to/image.jpg --device 0