# Emotion estimation from video footage with LSTM

## TL;DR

This paper introduces a new LSTM model for estimating emotions from video footage using facial expressions, achieving good accuracy with lower computational costs.

## Contribution

BlendFER-Lite is a novel LSTM model that uses MediaPipe Blendshapes for emotion estimation with reduced computational costs.

## Key findings

- BlendFER-Lite achieves 71% accuracy on the FER2013 dataset.
- The model has an F1-score of 62%, meeting the dataset's accuracy benchmark.
- It significantly reduces computational costs compared to existing methods.

## Abstract

Emotion estimation is a field that has been studied for a long time, and several approaches using machine learning models exist. This article presents BlendFER-Lite, an LSTM model that uses Blendshapes from the MediaPipe library to analyze facial expressions detected from a live-streamed camera feed. This model is trained on the FER2013 dataset and achieves 71% accuracy and an F1-score of 62%, meeting the accuracy benchmark for the FER2013 dataset while significantly reducing computational costs compared to current methods. For the sake of reproducibility, the code repository, datasets, and models proposed in this paper, in addition to the preprint, can be found on Hugging Face at: https://huggingface.co/papers/2501.13432.

D8, H51

35A01, 65L10, 65L12, 65L20, 65L70