Face Feature Module Usage Tutorial¶
I. Overview¶
Face feature models typically take standardized face images processed through detection, extraction, and keypoint correction as input. These models extract highly discriminative facial features from these images for subsequent modules, such as face matching and verification tasks.
II. Supported Model List¶
👉Details of Model List
| Model | Model Download Link | Output Feature Dimension | Acc (%) AgeDB-30/CFP-FP/LFW |
GPU Inference Time (ms) | CPU Inference Time (ms) | Model Size (M) | Description |
|---|---|---|---|---|---|---|---|
| MobileFaceNet | Inference Model/Trained Model | 128 | 96.28/96.71/99.58 | 5.7 | 101.6 | 4.1 | Face feature model trained on MobileFaceNet with MS1Mv3 dataset |
| ResNet50_face | Inference Model/Trained Model | 512 | 98.12/98.56/99.77 | 8.7 | 200.7 | 87.2 | Face feature model trained on ResNet50 with MS1Mv3 dataset |
Note: The above accuracy metrics are Accuracy scores measured on the AgeDB-30, CFP-FP, and LFW datasets, respectively. All model GPU inference times are based on an NVIDIA Tesla T4 machine with FP32 precision. CPU inference speeds are based on an Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz with 8 threads and FP32 precision.
III. Quick Integration¶
❗ Before quick integration, please install the PaddleX wheel package. For details, refer to the PaddleX Local Installation Tutorial
After installing the whl package, a few lines of code can complete the inference of the face feature module. You can switch models under this module freely, and you can also integrate the model inference of the face feature module into your project. Before running the following code, please download the example image to your local machine.
from paddlex import create_model
model_name = "MobileFaceNet"
model = create_model(model_name)
output = model.predict("face_recognition_001.jpg", batch_size=1)
for res in output:
res.print(json_format=False)
res.save_to_json("./output/res.json")
For more information on using the PaddleX single-model inference API, refer to the PaddleX Single Model Python Script Usage Instructions.
IV. Custom Development¶
If you aim for higher accuracy with existing models, you can leverage PaddleX's custom development capabilities to develop better face feature models. Before developing face feature models with PaddleX, ensure you have installed the PaddleX PaddleClas plugin. The installation process can be found in the PaddleX Local Installation Tutorial
4.1 Data Preparation¶
Before model training, you need to prepare the dataset for the corresponding task module. PaddleX provides data validation functionality for each module, and only data that passes validation can be used for model training. Additionally, PaddleX provides demo datasets for each module, allowing you to complete subsequent development based on the official demo data. If you wish to use a private dataset for subsequent model training, the training dataset for the face feature module is organized in a general image classification dataset format. You can refer to the PaddleX Image Classification Task Module Data Annotation Tutorial. If you wish to use a private dataset for subsequent model evaluation, note that the validation dataset format for the face feature module differs from the training dataset format. Please refer to Section 4.1.4 Data Organization Face Feature Module
4.1.1 Demo Data Download¶
You can use the following commands to download the demo dataset to a specified folder:
cd /path/to/paddlex
wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/face_rec_examples.tar -P ./dataset
tar -xf ./dataset/face_rec_examples.tar -C ./dataset/
4.1.2 Data Validation¶
A single command can complete data validation:
python main.py -c paddlex/configs/face_recognition/MobileFaceNet.yaml \
-o Global.mode=check_dataset \
-o Global.dataset_dir=./dataset/face_rec_examples
After executing the above command, PaddleX will validate the dataset and collect its basic information. Upon successful execution, the log will print the message Check dataset passed !. The validation result file will be saved in ./output/check_dataset_result.json, and related outputs will be saved in the ./output/check_dataset directory of the current directory. The output directory includes visualized example images and histograms of sample distributions.
👉 Validation Result Details (Click to Expand)
The specific content of the validation result file is:
{
"done_flag": true,
"check_pass": true,
"attributes": {
"train_label_file": "../../dataset/face_rec_examples/train/label.txt",
"train_num_classes": 995,
"train_samples": 1000,
"train_sample_paths": [
"check_dataset/demo_img/01378592.jpg",
"check_dataset/demo_img/04331410.jpg",
"check_dataset/demo_img/03485713.jpg",
"check_dataset/demo_img/02382123.jpg",
"check_dataset/demo_img/01722397.jpg",
"check_dataset/demo_img/02682349.jpg",
"check_dataset/demo_img/00272794.jpg",
"check_dataset/demo_img/03151987.jpg",
"check_dataset/demo_img/01725764.jpg",
"check_dataset/demo_img/02580369.jpg"
],
"val_label_file": "../../dataset/face_rec_examples/val/pair_label.txt",
"val_num_classes": 2,
"val_samples": 500,
"val_sample_paths": [
"check_dataset/demo_img/Don_Carcieri_0001.jpg",
"check_dataset/demo_img/Eric_Fehr_0001.jpg",
"check_dataset/demo_img/Harry_Kalas_0001.jpg",
"check_dataset/demo_img/Francis_Ford_Coppola_0001.jpg",
"check_dataset/demo_img/Amer_al-Saadi_0001.jpg",
"check_dataset/demo_img/Sergei_Ivanov_0001.jpg",
"check_dataset/demo_img/Erin_Runnion_0003.jpg",
"check_dataset/demo_img/Bill_Stapleton_0001.jpg",
"check_dataset/demo_img/Daniel_Bruehl_0001.jpg",
"check_dataset/demo_img/Clare_Short_0004.jpg"
]
},
"analysis": {},
"dataset_path": "./dataset/face_rec_examples",
"show_type": "image",
"dataset_type": "ClsDataset"
}
The verification results mentioned above indicate that check_pass being True means the dataset format meets the requirements. Details of other indicators are as follows:
attributes.train_num_classes: The number of classes in this training dataset is 995;attributes.val_num_classes: The number of classes in this validation dataset is 2;attributes.train_samples: The number of training samples in this dataset is 1000;attributes.val_samples: The number of validation samples in this dataset is 500;attributes.train_sample_paths: The list of relative paths to the visualization images of training samples in this dataset;
4.1.3 Dataset Format Conversion / Dataset Splitting (Optional)¶
After completing the data validation, you can convert the dataset format and re-split the training/validation ratio by modifying the configuration file or adding hyperparameters.
👉 Details on Format Conversion / Dataset Splitting (Click to Expand)
The face feature module does not support data format conversion or dataset splitting.
4.1.4 Data Organization for Face Feature Module¶
The format of the validation dataset for the face feature module differs from the training dataset. If you need to evaluate model accuracy on private data, please organize your dataset as follows:
face_rec_dataroot # Root directory of the dataset, the directory name can be changed
├── train # Directory for saving the training dataset, the directory name cannot be changed
├── images # Directory for saving images, the directory name can be changed but should correspond to the content in label.txt
├── xxx.jpg # Face image file
├── xxx.jpg # Face image file
...
└──label.txt # Training set annotation file, the file name cannot be changed. Each line gives the relative path of the image to `train` and the face image class (face identity) id, separated by a space. Example content: images/image_06765.jpg 0
├── val # Directory for saving the validation dataset, the directory name cannot be changed
├── images # Directory for saving images, the directory name can be changed but should correspond to the content in pair_label.txt
├── xxx.jpg # Face image file
├── xxx.jpg # Face image file
...
└── pair_label.txt # Validation dataset annotation file, the file name cannot be changed. Each line gives the paths of two images to be compared and a 0 or 1 label indicating whether the pair of images belong to the same person, separated by spaces.
Example content of the validation set annotation file pair_label.txt:
# Face image 1.jpg Face image 2.jpg Label (0 indicates the two face images do not belong to the same person, 1 indicates they do)
images/Angela_Merkel_0001.jpg images/Angela_Merkel_0002.jpg 1
images/Bruce_Gebhardt_0001.jpg images/Masao_Azuma_0001.jpg 0
images/Francis_Ford_Coppola_0001.jpg images/Francis_Ford_Coppola_0002.jpg 1
images/Jason_Kidd_0006.jpg images/Jason_Kidd_0008.jpg 1
images/Miyako_Miyazaki_0002.jpg images/Munir_Akram_0002.jpg 0
4.2 Model Training¶
Model training can be completed with a single command. Here is an example of training MobileFaceNet:
python main.py -c paddlex/configs/face_recognition/MobileFaceNet.yaml \
-o Global.mode=train \
-o Global.dataset_dir=./dataset/face_rec_examples
- Specify the path to the
.yamlconfiguration file for the model (here it isMobileFaceNet.yaml) - Specify the mode as model training:
-o Global.mode=train - Specify the path to the training dataset:
-o Global.dataset_dirOther related parameters can be set by modifying theGlobalandTrainfields in the.yamlconfiguration file or by appending parameters in the command line. For example, to specify the first two GPUs for training:-o Global.device=gpu:0,1; to set the number of training epochs to 10:-o Train.epochs_iters=10. For more modifiable parameters and their detailed explanations, refer to the configuration file instructions for the corresponding task module PaddleX Common Configuration Parameters for Model Tasks.
👉 More Details (Click to Expand)
- During model training, PaddleX automatically saves model weight files, defaulting to
output. To specify a save path, use the-o Global.outputfield in the configuration file. - PaddleX shields you from the concepts of dynamic graph weights and static graph weights. During model training, both dynamic and static graph weights are produced, and static graph weights are selected by default for model inference.
- When training other models, specify the corresponding configuration file. The correspondence between models and configuration files can be found in the PaddleX Model List (CPU/GPU).
After completing model training, all outputs are saved in the specified output directory (default is
./output/). Typically, the following outputs are included: train_result.json: A file that records the training results, indicating whether the training task was successfully completed, and includes metrics, paths to related files, etc.train.log: A log file that records changes in model metrics, loss variations, and other details during the training process.config.yaml: A configuration file that logs the hyperparameter settings for the current training session..pdparams,.pdema,.pdopt.pdstate,.pdiparams,.pdmodel: Files related to model weights, including network parameters, optimizer, EMA (Exponential Moving Average), static graph network parameters, and static graph network structure.
python main.py -c paddlex/configs/face_detection/MobileFaceNet.yaml \
-o Global.mode=evaluate \
-o Global.dataset_dir=./dataset/face_rec_examples
👉 More Details (Click to Expand)
During model evaluation, the path to the model weights file needs to be specified. Each configuration file has a default weight save path built in. If you need to change it, you can set it by appending a command line parameter, such as `-o Evaluate.weight_path="./output/best_model/best_model/model.pdparams"`. After completing the model evaluation, an `evaluate_result.json` file will be produced, which records the evaluation results. Specifically, it records whether the evaluation task was completed normally and the model's evaluation metrics, including Accuracy.python main.py -c paddlex/configs/face_recognition/MobileFaceNet.yaml \
-o Global.mode=predict \
-o Predict.model_dir="./output/best_model/inference" \
-o Predict.input="face_recognition_001.jpg"