Project on predicting diving scores and servicing the results with RAG
In sports, biased judgments by referees have long been a source of controversy.
To address this issue objectively, we propose the development of an AI-driven service that evaluates the performance of actions during games based on a rule-based system.
This service aims to provide users not only with the AI model’s evaluation results but also with a detailed explanation of the scores assigned.
By offering transparency in the decision-making process, this tool will enable users to determine whether a referee’s judgment was genuinely biased or fair, thus fostering greater trust and accountability in sports officiating.
User Input: diving video link (YouTube), specify the desired time segment, the actual score
The service will then process the video through a series of steps:
- OCR for Difficulty Identification
- Video Analysis and Score Prediction
- RAG-based Q&A
Our RAG (Retrieval-Augmented Generation) system was implemented using chatgpt4o-mini and ChromaDB. The performance of the RAG system was evaluated using RAGAS, yielding the following results:
- Context Precision: 1.0000
- Context Recall: 0.5167
- Faithfulness: 0.6935
- Answer Relevancy: 0.5916
These metrics reflect the effectiveness and accuracy of our RAG system in generating relevant and faithful responses based on the retrieved context.
- Dependencies Installation
- For Backend, Activate .venv and
pip install -r Backend/requirements.txt - For NSAQA, Activate another conda env and
pip install -r NSAQA/requirements.txt
- For Backend, Activate .venv and
- Prerequisite:
npm --version==10.8.2 ffmpegInstallation: You have to installffmpegfor video precessing. (follow this: https://medium.com/@vladakuc/compile-opencv-4-7-0-with-ffmpeg-5-compiled-from-the-source-in-ubuntu-434a0bde0ab6)- API Keys: You need to set up OPENAI_API_KEY in '.env' file
- Activate Backend and Frontend module first
$uvicorn main:app --reload # in ./Backend
$npm run serve # in ./Frontend
- Run Inference Server module
$uvicorn app:app --host 0.0.0.0 --port 8090 # in ./NSAQA
| 김대솔 | 김보담 | 김진형 | 김채현 | 박수연 | 양인혜 |
|---|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |