|
|
--- |
|
|
|
|
|
|
|
|
{} |
|
|
--- |
|
|
|
|
|
# Facial-Expression-Recognition |
|
|
This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the FER 2013 and AffectNet dataset datasets. It achieves the following results on the evaluation set: |
|
|
Accuracy - 0.922 |
|
|
Loss - 0.213 |
|
|
|
|
|
|
|
|
|
|
|
### Model Description |
|
|
|
|
|
The vit-face-expression model is a Vision Transformer fine-tuned for the task of facial emotion recognition. |
|
|
|
|
|
It is trained on the FER2013and AffectNet datasets, which consist of facial images categorized into eight different emotions: |
|
|
-anger |
|
|
-contempt |
|
|
-sad |
|
|
-happy |
|
|
-neutral |
|
|
-disgust |
|
|
-fear |
|
|
-surprise |
|
|
|
|
|
## Model Details |
|
|
The model has been fine-tuned using the following hyperparameters: |
|
|
|
|
|
| Hyperparameter | Value | |
|
|
|-------------------------|------------| |
|
|
| Train Batch Size | 32 | |
|
|
| Eval Batch Size | 64 | |
|
|
| Learning Rate | 2e-4 | |
|
|
| Gradient Accumulation | 2 | |
|
|
| LR Scheduler | Linear | |
|
|
| Warmup Ratio | 0.04 | |
|
|
| Num Epochs | 10 | |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
## How to Get Started with the Model |
|
|
Example usage: |
|
|
|
|
|
```python |
|
|
from transformers import AutoImageProcessor, AutoModelForImageClassification, pipeline |
|
|
|
|
|
pipe = pipeline("image-classification", model="HardlyHumans/Facial-expression-detection") |
|
|
|
|
|
processor = AutoImageProcessor.from_pretrained("HardlyHumans/Facial-expression-detection") |
|
|
model = AutoModelForImageClassification.from_pretrained("HardlyHumans/Facial-expression-detection") |
|
|
|
|
|
labels = model.config.id2label |
|
|
|
|
|
outputs = model(**inputs) |
|
|
|
|
|
logits = outputs.logits |
|
|
predicted_class_idx = logits.argmax(-1).item() |
|
|
predicted_label = labels[predicted_class_idx] |
|
|
``` |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
## Environmental Impact |
|
|
The net estimated CO2 emission using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) scale is around 8.82kg of CO2. |
|
|
|
|
|
|
|
|
|
|
|
- **Developed by:** Hardly Humans club, IIT Dharwad |
|
|
- **Model type:** Vision transformer |
|
|
- **License:** MIT |
|
|
- **Finetuned from model:** google/vit-base-patch16-224-in21k |
|
|
|
|
|
- **Hardware Type:** T4 |
|
|
- **Hours used:** 8+27 |
|
|
- **Cloud Provider:** Google collabotary service |
|
|
- **Compute Region:** South asia-1 |
|
|
- **Carbon Emitted:** 8.82 |
|
|
|
|
|
|
|
|
### Model Architecture and Objective |
|
|
|
|
|
|
