- Windows 10 laptop
- CPU i7-11375H
- Opencv 3.4.5
- OneDNN 2.50 (2021.4)
- resnet18 Model
- Function abstraction in an easy-to-use form
- Primitive function validation
- Weight loader
- Data loader
- Image preprocess
- Modeling
- Match all results with PyTorch
- Optimization
- Performace evaluation(Execution time of 100 iteration for one 224x224x3 image)
| Pytorch | OneDNN | OneDNN | OneDNN | |
| Description | general | Specific data type | Any data type | Any data type + fused post ops |
| Avg Duration time [ms] | 27.59 ms | 67.65 ms | 24.40 ms | 22.81 ms |
| FPS [frame/sec] | 36 fps | 15 fps | 41 fps | 44 fps |
| File | Resnet18_py/inference.py | resnet18_v1.cpp | resnet18_v2.cpp | resnet18_v3.cpp |
- oneDNN github : https://github.com/oneapi-src/oneDNN#installation
- onednn v2.50 documentation :https://oneapi-src.github.io/oneDNN/
- download : https://www.intel.com/content/www/us/en/developer/tools/oneapi/onednn.html#gs.i239i5