From d7c7c087ab4a49b7f344a8fd289efdad0b6c6880 Mon Sep 17 00:00:00 2001
From: Tudosie Razvan <80643689+Mars-Zero@users.noreply.github.com>
Date: Sun, 15 Dec 2024 21:22:53 +0200
Subject: [PATCH] Update README.md

---
 README.md | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index db68bd5..5161a79 100644
--- a/README.md
+++ b/README.md
@@ -1,4 +1,12 @@
-# IA4 Project-> Real-Time Sign Language Translator
+### Real-Time Sign Language Translator
 
 
-Applied informatics project
+#AI Model
+A Google AI gesture recognition model was used for ASL image recognition. This model was trained on a 2GB curated dataset compiled from various Kaggle repositories.
+The model achieves an accuracy of 93.6% and a loss of 0.1563.
+In practice, the lighting of the environment significantly affects the accuracy of classification because the training data primarily consists of images with good to excellent lighting conditions.
+Other Hugging Face alternatives tested were slower and less accurate overall, based on our evaluations.
+
+#Validating AI Sign Language Results
+To validate results and generate clear text output, an OpenAI API call was used. With a custom prompt, the classification results are combined into a meaningful sentence.
+A Mistral alternative was tested but found to be less effective. The OpenAI solution provides clearer messages and better contextual guesses, resulting in sentences closer to the intended user translation.