Epoch	Training Loss	Validation Loss	Accuracy	Precision	Recall	F1
1	0.4788	0.6879	0.7249	0.6906	0.6889	0.6863
2	0.3865	0.6226	0.7400	0.7005	0.7124	0.7015
3	0.3277	0.6182	0.7442	0.7042	0.7136	0.7045
4	0.2713	0.6481	0.7400	0.7025	0.7112	0.7022
5	0.2372	0.7596	0.7544	0.7263	0.7214	0.7216
6	0.1916	0.7392	0.7554	0.7234	0.7241	0.7223
7	0.1569	0.8017	0.7495	0.7179	0.7150	0.7078
8	0.1297	0.8738	0.7615	0.7306	0.7220	0.7228
9	0.1129	0.9251	0.7567	0.7283	0.7194	0.7175
10	0.0987	0.8925	0.7653	0.7337	0.7263	0.7267
11	0.0819	1.0842	0.7616	0.7398	0.7168	0.7227
12	0.0671	0.9245	0.7620	0.7332	0.7277	0.7254
13	0.0591	0.9775	0.7670	0.7400	0.7277	0.7298
14	0.0523	1.0465	0.7664	0.7449	0.7257	0.7298
15	0.0476	1.0487	0.7571	0.7274	0.7219	0.7189
16	0.0481	1.1164	0.7619	0.7414	0.7188	0.7237
17	0.0378	1.1738	0.7567	0.7376	0.7139	0.7196

✅ 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

python

from transformers import BeitImageProcessor, BeitForImageClassification
from PIL import Image

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

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

Name: Yamini Devika Kanakam
Base model: microsoft/beit-large-patch16-224-pt22k-ft22k

📚 Citation

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

bibtex

@misc{kanakam2025beit_affectnet,
  title={BEiT-Large fine-tuned on AffectNet for Emotion Detection},
  author={Yamini Devika Kanakam},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/YaminiDevikaKanakam/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}
}