An innovative computer vision project that leverages the physical properties of aromatic smoke for advanced physical simulations and visual analysis.
SmokePhysAI combines computer vision techniques with fluid dynamics to analyze and simulate the behavior of aromatic smoke. This project enables:
- Real-time smoke pattern recognition
- Physical property extraction from smoke movements
- Predictive modeling of smoke dispersion
- Visual enhancement of smoke dynamics
- 🌀 Smoke Tracking: Advanced particle tracking algorithms
- 🔍 Property Analysis: Density, velocity, and turbulence measurement
- 🤖 AI Simulation: Predictive modeling using physics-informed neural networks
- 📊 Data Visualization: Interactive 3D smoke visualization tools
Benchmark results comparing SmokePhysAI with traditional computer vision methods:
| Model | MSE | Physics Correlation | Inference Time (ms) |
|---|---|---|---|
| SmokePhysAI | 0.002955 | 0.9957 | 610.92 |
| Farneback | 0.699607 | N/A | 3.98 |
| Lucas-Kanade | 0.723172 | N/A | 0.71 |
Notes:
- MSE (Mean Squared Error) measures reconstruction accuracy (lower is better)
- Physics Correlation measures accuracy of physical property prediction (1.0 is perfect)
- Inference Time is per-frame processing time (lower is better)
SmokePhysAI achieves 200x higher accuracy than traditional methods while capturing physical properties with near-perfect correlation. We're actively working to optimize inference speed.
git clone https://github.com/MengAiDev/SmokePhysAI.git
cd SmokePhysAI
pip install -e .You can download pre-trained models from ModelScope:
Train from scratch:
python train.py --config config/config.yamlRun inference:
python inference.py --model_path model.pth --config config/config.yamlRun benchmark:
python benchmark.py --checkpoint model.pth --num_samples 100Please see the inference_output/ directory for the results, which trained by myself on RTX 3090.
Contributions are welcome!
This project is licensed under the Apache 2.0 License - see LICENSE for details.