This repository contains the code for fine-tuning the EfficientNet-B0 and EfficientNet-B2 models on the RAF-DB dataset. The streamlit app was created to test the fine-tuned models on custom images and videos. This project serves as our final project in the course CS331 - Advanced Computer Vision at UIT
This is how the streamlit app looks like. There are various options for face detectors, confidence thresholds and number of frames skipped (when working with input video), etc. that you can adjust to play around with the fine-tuned EfficientNet models.
| Input Image | Input Video |
|---|---|
 |
 |
Note that our pretrained EfficientNet model weights were obtained from this repository by Andrey V.Savchenko.
| Name | Student Id | Github |
|---|---|---|
| Tran Xuan Thanh | 21520456 | https://github.com/LukasAbraham |
| Mai Anh Quan | 21520411 | https://github.com/maqnitude |