BEiT-Large for Emotion Detection on FER-2013, RAF-DB, and AffectNet

This model is fine-tuned from microsoft/beit-large-patch16-224-pt22k-ft22k for facial emotion recognition using a combined dataset made from:

• FER-2013
• RAF-DB
• AffectNet (Cleaned & Balanced)

This fusion of diverse facial expression datasets improves the model's ability to generalize across real-world facial emotion recognition tasks.

🧠 Classes

The model predicts the following 7 basic emotion classes:

• 😠 anger
• 🤢 disgust
• 😨 fear
• 😀 happy
• 😐 neutral
• 😢 sad
• 😲 surprise

📈 Training Metrics

✅ Final Accuracy: ~76.2%
✅ Final F1 Score: ~0.725

🛠️ Training Configuration

The model was trained using the Hugging Face Trainer with the following key arguments:

• num_train_epochs=20
• per_device_train_batch_size=32
• per_device_eval_batch_size=32
• gradient_accumulation_steps=2
• learning_rate=2e-5
• fp16=True (mixed precision)
• eval_strategy="epoch"
• save_strategy="epoch"
• save_total_limit=2
• load_best_model_at_end=True
• metric_for_best_model="f1"
• greater_is_better=True

🔄 Confusion Matrix

📊 Classification Report

🔧 How to Use

from transformers import BeitImageProcessor, BeitForImageClassification
from PIL import Image

image = Image.open("/your/image/path.jpg").convert("RGB")

processor = BeitImageProcessor.from_pretrained("Tanneru/Facial-Emotion-Detection-FER-RAFDB-AffectNet-BEIT-Large")
model = BeitForImageClassification.from_pretrained("Tanneru/Facial-Emotion-Detection-FER-RAFDB-AffectNet-BEIT-Large")

inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)
predicted_class = outputs.logits.argmax(-1).item()

print("Predicted emotion:", model.config.id2label[predicted_class])

📄 License

This model is released under the Apache 2.0 License. You are free to use, modify, and distribute the model with attribution.

✍️ Author

• Username: Tanneru
• Base model: microsoft/beit-large-patch16-224-pt22k-ft22k

📚 Citation

If you use this model in your work, please cite:

@misc{tanneru2025beit_affectnet,
  title={BEiT-Large fine-tuned on AffectNet for Emotion Detection},
  author={Tanneru},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/Tanneru/Facial-Emotion-Detection-FER-RAFDB-AffectNet-BEIT-Large}},
}

@article{bao2021beit,
  author       = {Hangbo Bao and Li Dong and Furu Wei},
  title        = {BEiT: BERT Pre-Training of Image Transformers},
  journal      = {CoRR},
  volume       = {abs/2106.08254},
  year         = {2021},
  url          = {https://arxiv.org/abs/2106.08254},
  archivePrefix = {arXiv},
  eprint       = {2106.08254},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}