Files
insightface/recognition/tools/README.md
2021-03-04 17:52:51 +08:00

67 lines
3.0 KiB
Markdown

## I. CPP-Align
-
## II. Face Mask Renderer
We provide a simple tool to add masks on face images automatically.
We can use this tool to do data augmentation while training our face recognition models.
| Face Image | OP | Mask Image | Out |
| ------- | ------ | --------- | ----------- |
| <img src="https://github.com/deepinsight/insightface/blob/master/deploy/Tom_Hanks_54745.png" alt="face" height="112" /> | +F | <img src="https://github.com/nttstar/insightface-resources/blob/master/images/mask1.jpg" alt="mask" height="112" /> | <img src="https://github.com/nttstar/insightface-resources/blob/master/images/mask_out1.jpg?raw=true" alt="mask" height="112" /> |
| <img src="https://github.com/deepinsight/insightface/blob/master/deploy/Tom_Hanks_54745.png" alt="face" height="112" /> | +F | <img src="https://github.com/nttstar/insightface-resources/blob/master/images/black-mask.png" alt="mask" height="112" /> | <img src="https://github.com/nttstar/insightface-resources/blob/master/images/mask_out3.jpg?raw=true" alt="mask" height="112" /> |
| <img src="https://github.com/deepinsight/insightface/blob/master/deploy/Tom_Hanks_54745.png" alt="face" height="112" /> | +H | <img src="https://github.com/nttstar/insightface-resources/blob/master/images/mask2.jpg?raw=true" alt="mask" height="112" /> | <img src="https://github.com/nttstar/insightface-resources/blob/master/images/mask_out2h.jpg?raw=true" alt="mask" height="112" /> |
**F** means FULL while **H** means HALF.
### Data Prepare
- face3d library
Please copy and install [`face3d`](https://github.com/YadiraF/face3d) library and put it under `face3d/` directory. Then make sure you can `import face3d` in python env.
- BFM models
Please follow the tutorial of [https://github.com/YadiraF/face3d/tree/master/examples/Data/BFM](https://github.com/YadiraF/face3d/tree/master/examples/Data/BFM) to generate `BFM.mat` and `BFM_UV.mat`.
- 3D68 model
Download our 3D68 pretrained model(if1k3d68-0000.params) from [baiducloud](https://pan.baidu.com/s/1f9UtTpaW4l65DMRZ5wmJew)(passwd:2zmz) or [googledrive](https://drive.google.com/file/d/1Tv9f6Iqm4423qEDdDwbyjk00plEfeG8a/view?usp=sharing) and put it under `assets_mask/`. Please also put the above BFM data model into this directory.
- mask images
You can get example mask images by checking the image url from the table above.
### Add Mask to Face Image
Please refer to `make_renderer.py` for detail example.
(1) init renderer:
```
tool = MaskRenderer('./assets_mask')
```
(2) load face and mask images
```
image = cv2.imread("../../deploy/Tom_Hanks_54745.png")
mask_image = cv2.imread("masks/mask2.jpg")
```
(3) build necessary params for face image, this can be done in offline.
```
params = tool.build_params(image)
```
(4) do mask render, it costs about `10ms` on 224x224 UV size, CPU single thread.
```
mask_out = tool.render_mask(image, mask_image, params)
```
(5) do half mask render.
```
mask_half_out = tool.render_mask(image, mask_image, params, positions=[0.1, 0.5, 0.9, 0.7])
```