Skip to content

Latest commit

 

History

History

model_converter

README.md

Insightface PyTorch To Tensorflow Lite (.tflite)

Recent Update

2021-05-20: First commit.

Content

Insightface PyTorch model (.pth) convert to Tensorflow Lite model (.tflite)

Set Up

Python 3.6
PyTorch 1.7

CPU Version

Package Version
traceback2 1.4.0
numpy 1.19.5
opencv-python 4.5.2.52
torchsummary 1.5.1
pkg-resources1 0.0.7
scipy 1.5.4
onnx-tf 1.6.0
onnxruntime 1.7.0
tensorflow 2.4.1
tensorflow_addons 0.12.1

GPU Version (CUDA 10.1)

Package Version
traceback2 1.4.0
numpy 1.19.5
opencv-python 4.5.2.52
torchsummary 1.5.1
pkg-resources1 0.0.7
scipy 1.5.4
onnx-tf 1.6.0
onnxruntime 1.7.0
tensorflow-gpu 2.4.1
tensorflow_addons 0.12.1

Set up environment

Create Python Environment

Python 3.6

conda create -n insightface python=3.6

Install PyTorch

v1.7.1 - Linux and Windows

# CUDA 11.0
pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

# CUDA 10.2
pip install torch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2

# CUDA 10.1
pip install torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

# CUDA 9.2
pip install torch==1.7.1+cu92 torchvision==0.8.2+cu92 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

# CPU only
pip install torch==1.7.1+cpu torchvision==0.8.2+cpu torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

More details see PyTorch Website in docs.

Install Python Package

CPU Version

requirements.txt

pip install -r requirements.txt

GPU Version (CUDA 10.1)

requirements-gpu.txt

pip install -r requirements-gpu.txt

PyTorch To Tensorflow Lite

usage: pytorch_to_tflite.py [-h] [--pytorch_model_path PYTORCH_MODEL_PATH]
                            [--pytorch_weight_path PYTORCH_WEIGHT_PATH]
                            [--input_shape INPUT_SHAPE]
                            [--input_names INPUT_NAMES]
                            [--output_names OUTPUT_NAMES]
                            [--tensorflow_lite_model_output_path TENSORFLOW_LITE_MODEL_OUTPUT_PATH]

PyTorch model convert to tflite model

optional arguments:
  -h, --help            show this help message and exit
  --pytorch_model_path PYTORCH_MODEL_PATH
                        where is the pytorch model.
  --pytorch_weight_path PYTORCH_WEIGHT_PATH
                        where is the pytorch weight.
  --input_shape INPUT_SHAPE
                        input shape for pytorch model. ex. 3,112,112
  --input_names INPUT_NAMES
                        the input names to use for onnx model. ex. input
  --output_names OUTPUT_NAMES
                        the output name to use for onnx model. ex. output
  --tensorflow_lite_model_output_path TENSORFLOW_LITE_MODEL_OUTPUT_PATH
                        where is save to tflite model.


python pytorch_to_tflite.py --pytorch_model_path "../../../../Deep_Learning/InsightFace/Python/Models/PyTorch/emore_mask_resnet_r18/6_backbone.pth" --input_shape 3,112,112 --input_names input --output_names output --tensorflow_lite_model_output_path "../../../../Deep_Learning/InsightFace/Python/Models/PyTorch/emore_mask_resnet_r18/emore_mask_r18.tflite"