github.com/Tamminhdiep97/model_hub ↗
This is where i save interested model, convert and test time inference with onnxruntime
Open this visualization on its own page →
Contributors
1
Lines of Code
3
From
2022-04-07
To
2022-04-07
About Tamminhdiep97/model_hub
model_hub is a Python project focused on face recognition models, serving as a collection for saving, converting, and benchmarking neural networks. The repository provides tools to convert PyTorch models to ONNX format and measure inference performance using ONNX Runtime, allowing users to easily test and compare different model architectures.
The project includes experimental work on optimizing the IR-50 face recognition network by replacing its final layer with a Ghostnet layer, creating an IR-GHOST-50 variant. This modification achieved significant parameter reduction of approximately 10 million parameters while cutting average inference time roughly in half on test inputs. The repository documents training experiments on augmented and masked versions of the Glint360k dataset, with results tracked and visualized through Tensorboard.
Users can configure and run tests through a config.py file to evaluate different models' performance characteristics. The work demonstrates a practical approach to model optimization for face recognition tasks, balancing network efficiency with inference speed through architectural modifications rather than simple pruning or quantization approaches.