# lightweight_openpose_tensorflow

**Repository Path**: LuckyHouse/lightweight_openpose_tensorflow

## Basic Information

- **Project Name**: lightweight_openpose_tensorflow
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-06-15
- **Last Updated**: 2023-12-06

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

**Lightweight OpenPose implemented in TensorFlow 2.0**

This repository provides a clean implementation of [Lightweight OpenpPose](https://arxiv.org/pdf/1811.12004.pdf)
in TensorFlow 2. I try to implement every possible thing in OOP. Hope it helps.
You could find the original implementation in PyTorch [here](https://github.com/Daniil-Osokin/lightweight-human-pose-estimation.pytorch)

***Key Features***

- [x] Dataset Loader of COCO2017 using `tf.data` input pipeline
- [x] Custom training loop using `tf.GradientTape`
- [x] Custom model implementation using `tf.keras`
- [x] Evaluation code which uses COCO format
- [x] Various exporting options to SavedModel, FrozenGraph, TFLite, MNN
- [x] Inference code for each exported format
- [x] `tf.function` and `Concrete Function`

***Usage***

****Training****
```bash
python script/train.py configs/train.yaml
```

****Evaluating****

```bash
python script/val.py configs/val.yaml
```

****Export****

```bash
python utils/export.py configs/export.yaml
```

****Inference****
```bash
python serving/native configs/val.yaml
```

***Note:***
If you have any question, feel free to open an issues or reach me out at this [email](minhhoangbui.vn@gmail.com)