Repository navigation
·
1101 commits
to staging
since this release
Visual Cheat Detection System for Counter-Strike 2
Initial alpha release featuring AI-powered cheat detection using Vision Transformer
technology.
Key Features
- Automated clip extraction - Detects killshots from gameplay audio and creates 2-second
clips - Local model training - Train custom models on your labeled footage (cheater/legitimate)
- Deep learning analysis - 1B parameter Vision Transformer analyzes frames 85-100 of each
clip - Confidence scoring - Detailed results from 0.0 (likely legitimate) to 1.0 (likely
cheating)
Technical Details
- Vision Transformer (VideoMAE v2) architecture
- Resolution-adaptive processing
- Frame-by-frame analysis with visual output
- Flask web interface at localhost:5000
- CUDA GPU support required
Installation
- Windows: WSL-based installation with provided .bat scripts
- Linux: Direct installation via install.sh
- Model: Download required 1.9GB base model from HuggingFace
Alpha Notes
- This is early alpha software for testing and feedback
- Requires minimum 48 clips for training
- Manual removal of spectator/proximity kills needed
- Results are guidance, not definitive proof
Resources
- Full documentation in README
---Version: 1.0.0-alpha