FaceRecognitionDotNet

https://github.com/takuya-takeuchi/facerecognitiondotnet
Admin
A .NET port of the world's simplest facial recognition API, enabling cross-platform facial...
Tokens:21,126
Snippets:192
Trust Score:8.7
License:MIT
Update:4 months ago
Show doc for...
Tokens:
Raw
### Install FaceRecognition.NET on Linux (Ubuntu/Debian)

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Installs necessary libraries for FaceRecognition.NET on Ubuntu/Debian-based Linux distributions.

```bash
sudo apt-get install libx11-6 libopenblas-dev libgdiplus
```

--------------------------------

### Run Example with Batch Size

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/FindFacesInBatches/README.md

Example command to run the FindFacesInBatches program with a batch size of 32. The output shows the number of faces found per frame and their pixel locations.

```powershell
$ pwsh Run.ps1 32
I found 0 face(s) in frame #0.
I found 0 face(s) in frame #1.
I found 0 face(s) in frame #2.
I found 0 face(s) in frame #3.
I found 0 face(s) in frame #4.
I found 0 face(s) in frame #5.
I found 0 face(s) in frame #6.
...
...
...
I found 3 face(s) in frame #250.
 - A face is located at pixel location Top: 50, Left: 312, Bottom: 145, Right: 407
 - A face is located at pixel location Top: 100, Left: 380, Bottom: 179, Right: 459
 - A face is located at pixel location Top: 92, Left: 156, Bottom: 171, Right: 235
I found 3 face(s) in frame #251.
 - A face is located at pixel location Top: 50, Left: 312, Bottom: 145, Right: 407
 - A face is located at pixel location Top: 100, Left: 380, Bottom: 179, Right: 459
 - A face is located at pixel location Top: 92, Left: 164, Bottom: 171, Right: 243
I found 3 face(s) in frame #252.
 - A face is located at pixel location Top: 50, Left: 312, Bottom: 145, Right: 407
 - A face is located at pixel location Top: 92, Left: 156, Bottom: 171, Right: 235
 - A face is located at pixel location Top: 100, Left: 380, Bottom: 179, Right: 459
I found 3 face(s) in frame #253.
 - A face is located at pixel location Top: 50, Left: 312, Bottom: 145, Right: 407
 - A face is located at pixel location Top: 92, Left: 156, Bottom: 171, Right: 235
 - A face is located at pixel location Top: 100, Left: 380, Bottom: 179, Right: 459
I found 3 face(s) in frame #254.
 - A face is located at pixel location Top: 50, Left: 312, Bottom: 145, Right: 407
 - A face is located at pixel location Top: 92, Left: 156, Bottom: 171, Right: 235
 - A face is located at pixel location Top: 100, Left: 380, Bottom: 179, Right: 459
I found 3 face(s) in frame #255.
 - A face is located at pixel location Top: 50, Left: 312, Bottom: 145, Right: 407
 - A face is located at pixel location Top: 92, Left: 164, Bottom: 171, Right: 243
 - A face is located at pixel location Top: 100, Left: 380, Bottom: 179, Right: 459
```

--------------------------------

### Run Face Recognition Encoding Example

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/Encoding/README.md

Command to run the encoding example, specifying the model and image paths. This will output bounding box coordinates and the face encoding as binary data.

```bash
$ dotnet run -c Release -- --model models --image obama-240p.jpg
l: 190, t: 32, r: 252, b: 94
-0.0858781188726425, 0.108835116028786, 0.0365724340081215, -0.0589281246066093, 0.0132484100759029, -0.00542659685015678, -0.065446712076664, -0.110336743295193, 0.19452229142189, -0.0979864299297333, 0.2352324873209, 0.0787549242377281, -0.213130414485931, -0.163164556026459, 0.0599864907562733, 0.114194087684155, -0.174584299325943, -0.0919561088085175, -0.0930094122886658, -0.117708653211594, 0.0325165763497353, -0.00997351761907339, 0.106906838715076, 0.0505531094968319, -0.144327610731125, -0.340307205915451, -0.0607180781662464, -0.188485234975815, -0.010340285487473, -0.108085952699184, -0.0919387713074684, 0.00310934125445783, -0.180108249187469, -0.124415650963783, 0.0167518854141235, -0.0189712662249804, 0.0118233505636454, -0.00810264237225056, 0.211996838450432, 0.026537474244833, -0.130677983164787, 0.0932836830615997, 0.0186070054769516, 0.207101136445999, 0.277875900268555, 0.0787801891565323, -0.00383740104734898, -0.0775237083435059, 0.126798510551453, -0.228157266974449, 0.0655233263969421, 0.155302807688713, 0.0920143350958824, 0.0186425745487213, 0.0837418735027313, -0.191329553723335, -0.00609584245830774, 0.0760508179664612, -0.14967368543148, 0.030364228412509, 0.0180820152163506, -0.0949648693203926, -0.0386509634554386, 0.0592935085296631, 0.182954981923103, 0.107051037251949, -0.113879278302193, -0.0532004535198212, 0.131714105606079, -0.0395184531807899, 0.015960929915309, 0.0232246778905392, -0.193916633725166, -0.200455144047737, -0.232742950320244, 0.0793688371777534, 0.374781519174576, 0.173277884721756, -0.211374044418335, -9.25520434975624E-05, -0.21218629181385, 0.0490190535783768, 0.0651776567101479, 0.015591973438859, -0.0640056058764458, -0.136019349098206, -0.0460829176008701, 0.0354794897139072, 0.0894847884774208, 0.0356751903891563, -0.0501493290066719, 0.225857928395271, -0.0144199710339308, 0.0671800822019577, 0.0112739009782672, 0.0495598912239075, -0.14527502655983, -0.0147501565515995, -0.158632963895798, -0.0530679523944855, 0.0203705467283726, -0.030452735722065, 0.0460555553436279, 0.145609840750694, -0.250953495502472, 0.0573373362421989, 0.00578311737626791, -0.0420315787196159, 0.0314476825296879, 0.0746812149882317, -0.0375748537480831, -0.0473008044064045, 0.0588102079927921, -0.238515302538872, 0.237175717949867, 0.259294271469116, 0.0266373176127672, 0.170270383358002, 0.0662055835127831, 0.0214092433452606, -0.000137977302074432, -0.0245289821177721, -0.163018763065338, -0.0655205324292183, 0.0550495907664299, 0.0647143721580505, 0.0707048177719116, 0.00707003846764565
```

--------------------------------

### Install FaceRecognition.NET on Linux (RHEL/Fedora/CentOS)

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Installs necessary development packages for FaceRecognition.NET on RHEL-based Linux distributions.

```bash
sudo yum install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
sudo yum install libX11-devel openblas-devel libgdiplus
```

--------------------------------

### Install libgdiplus on macOS

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Installs the libgdiplus library on macOS using Homebrew.

```bash
brew install mono-libgdiplus
```

--------------------------------

### Training Progress Output

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/EmotionTraining/README.md

Example output showing the training progress, including epoch number, learning rate, average loss, and accuracy metrics for both training and testing datasets.

```text
Dataset: <dataset_dir>
              Epoch: 3000
      Learning Rate: 0.0005
  Min Learning Rate: 5E-09
     Min Batch Size: 512
Validation Interval: 30
           Use Mean: False
             Output: <dataset_dir>\result

Start load train images
Use Cache <dataset_dir>\train_cache.dat
Load train images: 29568
Start load test images
Use Cache <dataset_dir>\test_cache.dat
Load test images: 3286

step#: 0     learning rate: 0.0005  average loss: 0            steps without apparent progress: 0
Epoch: 0, learning Rate: 0.0005, average loss: 4.11948018428823
Epoch: 1, learning Rate: 0.0005, average loss: 3.1305690457948
Epoch: 2, learning Rate: 0.0005, average loss: 2.69736696763001
Epoch: 3, learning Rate: 0.0005, average loss: 2.45950934526035
Epoch: 4, learning Rate: 0.0005, average loss: 2.30691530587521
Epoch: 5, learning Rate: 0.0005, average loss: 2.19935585556312
Epoch: 6, learning Rate: 0.0005, average loss: 2.11859311641417
Epoch: 7, learning Rate: 0.0005, average loss: 2.0555247708289
Epoch: 8, learning Rate: 0.0005, average loss: 2.00418632351514
Epoch: 9, learning Rate: 0.0005, average loss: 1.96132853099646
Epoch: 10, learning Rate: 0.0005, average loss: 1.92488743458344
Epoch: 11, learning Rate: 0.0005, average loss: 1.89341635029666
Epoch: 12, learning Rate: 0.0005, average loss: 1.86622101266595
Epoch: 13, learning Rate: 0.0005, average loss: 1.84257901432909
Epoch: 14, learning Rate: 0.0005, average loss: 1.82172028172718
Epoch: 15, learning Rate: 0.0005, average loss: 1.80331000603231
Epoch: 16, learning Rate: 0.0005, average loss: 1.78676114864211

Epoch: 384, learning Rate: 5E-09, average loss: 1.39896950052756
Saved state to Corrective_re-annotation_of_FER_CK+_KDEF-mlp_3000_0.0005_5E-09_512
Epoch: 385, learning Rate: 5E-09, average loss: 1.39896063521396
Epoch: 386, learning Rate: 5E-10, average loss: 1.39897608550845
Saved state to Corrective_re-annotation_of_FER_CK+_KDEF-mlp_3000_0.0005_5E-09_512_
done training
training num_right: 14257
training num_wrong: 15311
training accuracy:  0.482176677489177
testing num_right: 1703
testing num_wrong: 1583
testing accuracy:  0.518259281801582                    00:00:27
───────────────────────────────────────────────────────────────────
100.00% Step 4000 of 4000                                  00:00:00
───────────────────────────────────────────────────────────────────
training num_right: 4000
training num_wrong: 0
 training accuracy:  1.0000
100.00% Step 1000 of 1000                                  00:00:00
───────────────────────────────────────────────────────────────────
testing num_right: 1000
testing num_wrong: 0
 testing accuracy:  1.0000
train accuracy: 1.0000, test accuracy: 1.0000
```

--------------------------------

### Start Training Process

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/EmotionTrainingV2/README.md

Initiates the training process for the face emotion estimation model. This includes loading images, setting training parameters like epochs and learning rate, and saving results.

```log
cd <EmotionTraining_dir>
2022-02-06 17:27:34.0383 [INFO ]             Dataset: <EmotionTraining_dir> 
2022-02-06 17:27:34.2450 [INFO ]               Epoch: 3000 
2022-02-06 17:27:34.2450 [INFO ]       Learning Rate: 0.001 
2022-02-06 17:27:34.2450 [INFO ]   Min Learning Rate: 5E-09 
2022-02-06 17:27:34.2450 [INFO ]      Min Batch Size: 192 
2022-02-06 17:27:34.2450 [INFO ] Validation Interval: 1 
2022-02-06 17:27:34.2450 [INFO ]            Use Mean: False 
2022-02-06 17:27:34.2450 [INFO ]              Output: <EmotionTraining_dir>\result 
2022-02-06 17:27:34.2527 [INFO ]  
2022-02-06 17:27:34.2527 [INFO ] Start load train images 
2022-02-06 17:32:22.9455 [INFO ]             Dataset: <EmotionTraining_dir> 
2022-02-06 17:32:23.1567 [INFO ]               Epoch: 3000 
2022-02-06 17:32:23.1581 [INFO ]       Learning Rate: 0.001 
2022-02-06 17:32:23.1581 [INFO ]   Min Learning Rate: 5E-09 
2022-02-06 17:32:23.1581 [INFO ]      Min Batch Size: 192 
2022-02-06 17:32:23.1581 [INFO ] Validation Interval: 1 
2022-02-06 17:32:23.1581 [INFO ]            Use Mean: False 
2022-02-06 17:32:23.1581 [INFO ]              Output: <EmotionTraining_dir>\result 
2022-02-06 17:32:23.1581 [INFO ]  
2022-02-06 17:32:23.1581 [INFO ] Start load train images 
2022-02-06 17:32:49.4177 [INFO ] Load train images: 29568 
2022-02-06 17:32:49.4177 [INFO ] Start load test images 
2022-02-06 17:32:52.1883 [INFO ] Load test images: 3286 
2022-02-06 17:32:52.1883 [INFO ]  
2022-02-06 17:37:33.2620 [INFO ] Epoch: 0, learning Rate: 0.001, average loss: 1.83685714839136 
2022-02-06 17:38:13.9779 [INFO ] training num_right: 8214 
2022-02-06 17:38:13.9779 [INFO ] training num_wrong: 21354 

...

2022-02-06 23:22:20.0030 [INFO ] Epoch: 55, train accuracy: 0.996617965367965, test accuracy: 0.674071819841753 
2022-02-06 23:22:30.4702 [INFO ] Best Accuracy Model file is saved for train [0.996617965367965] 
2022-02-06 23:27:11.5971 [INFO ] Epoch: 56, learning Rate: 0.0001, average loss: 0.0236947398526919 
2022-02-06 23:27:52.3024 [INFO ] training num_right: 29460 
2022-02-06 23:27:52.3024 [INFO ] training num_wrong: 108 
2022-02-06 23:27:52.3024 [INFO ] training accuracy:  0.996347402597403 
2022-02-06 23:27:56.4852 [INFO ] testing num_right: 2231 
2022-02-06 23:27:56.4852 [INFO ] testing num_wrong: 1055 
2022-02-06 23:27:56.4852 [INFO ] testing accuracy:  0.678940961655508 
2022-02-06 23:28:25.0470 [INFO ]              Output: <EmotionTraining_dir>\result 
2022-02-06 23:28:25.0945 [INFO ]  
```

--------------------------------

### Run Face Detection Application

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/FaceDetection/README.md

Command to execute the built face detection application, specifying the directory containing images. Includes example image paths and bounding box coordinates.

```bash
$ dotnet run -c Release -- "-d=."
.\512px-President_Barack_Obama.jpg,79,314,203,189
.\Lenna.png,228,377,377,228
```

--------------------------------

### Load Image File

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Loads an image file into an Image object for processing. This example assumes the image file contains a single human face, but the library can handle images with multiple faces.

```csharp
using (Image imageA = FaceRecognition.LoadImageFile(imagePathA))
using (Image imageB = FaceRecognition.LoadImageFile(imagePathB))
```

--------------------------------

### Build the Project

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/EmotionTrainingV2/README.md

Build the project in Release configuration using the dotnet CLI. Ensure native DLLs are copied to the output directory.

```bash
dotnet build -c Release
```

--------------------------------

### Build and Run Face Recognition Sample

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/RgbBgr/README.md

Commands to build and run the sample application that demonstrates RGB vs. BGR image colorspace differences in face recognition. This includes adding the FaceRecognitionDotNet package and executing the build and run commands.

```bash
$ dotnet remove reference ../../src/FaceRecognitionDotNet\FaceRecognitionDotNet.csproj
$ dotnet add package FaceRecognitionDotNet
$ dotnet build -c Release
```

```bash
$ dotnet run -c Release
FaceLocations (Hog)
         File [LTRB]: 218, 219, 373, 374
          RGB [LTRB]: 218, 219, 373, 374
          BGR [LTRB]: 218, 219, 373, 374

FaceEncodings by File Location (Hog)
         vs  RGB [Distance]: 0
         vs  BGR [Distance]: 0.202584867566029
FaceEncodings by RGB Location (Hog)
         vs File [Distance]: 0
         vs  BGR [Distance]: 0.202584867566029
FaceEncodings by BGR Location (Hog)
         vs File [Distance]: 0.202584867566029
         vs  RGB [Distance]: 0.202584867566029

FaceLocations (Cnn)
         File [LTRB]: 198, 215, 340, 356
          RGB [LTRB]: 198, 215, 340, 356
          BGR [LTRB]: 212, 229, 354, 371

FaceEncodings by File Location (Cnn)
         vs  RGB [Distance]: 0
         vs  BGR [Distance]: 0.182024898919954
FaceEncodings by RGB Location (Cnn)
         vs File [Distance]: 0
         vs  BGR [Distance]: 0.182024898919954
FaceEncodings by BGR Location (Cnn)
         vs File [Distance]: 0.182024898919954
         vs  RGB [Distance]: 0.182024898919954
```

--------------------------------

### Build the Project

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/CustomClassificationDemo/README.md

Builds the custom classification demo project in Release configuration.

```bash
$ dotnet build -c Release
```

--------------------------------

### Run OpenCVSharp Sample

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/OpenCVSharpSample/README.md

Command to run the compiled Release build of the OpenCVSharp sample, specifying the directory containing model files.

```bash
$ dotnet run -c Release -- "-m=models"
Distance: 0.350636343769049 for obama-240p.jpg
Distance: 0.361864784857372 for obama-480p.jpg
Distance: 0.37696759951267 for obama-720p.jpg
Distance: 0.373414821495045 for obama-1080p.jpg
```

--------------------------------

### Run Demo with Trained Model

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/HelenTraining/README.md

Runs a demo using the trained model file, an input image, and the directory containing model files.

```bash
$ dotnet run -c Release -- demo -m "helen\helen-dataset.dat" -i "Steve_Jobs_Headshot_2010-CROP2.jpg" -d "models"
```

--------------------------------

### Run Head Pose Estimation Demo

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/HeadPoseEstimationDemo/README.md

Executes the Head Pose Estimation Demo in Release configuration. Ensure you are in the correct directory.

```bash
$ dotnet run -c Release
```

--------------------------------

### Create Dataset

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/EmotionTrainingV2/README.md

Generate train and test image list files from the 300W-LP dataset using the provided PowerShell script. The trainingRatio parameter controls the split between training and testing data.

```powershell
pwsh tools\GenerateTrainTestList.ps1 -rootDirectory <dataset_dir> \
                                     -trainingRatio 0.9
```

--------------------------------

### Create Dataset for Age Training

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/AgeTraining/README.md

PowerShell command to create a dataset from the Adience directory, splitting images into training and testing sets.

```powershell
pwsh tools\CreateDataset.ps1 -InputDirectory Adience -TrainRate 8 -OutputDirectory AdienceDataset -Max 0
```

--------------------------------

### Run Benchmark with Models

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/Benchmark/README.md

Command to execute the benchmark application, specifying the directory containing model files. This will output performance timings for various tasks at different resolutions.

```bash
$ dotnet run -c Release -- "-m=models"
```

--------------------------------

### Build FaceRecognitionDotNet Project

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/Encoding/README.md

Commands to remove project references, add the FaceRecognitionDotNet NuGet package, and build the project in Release configuration.

```bash
$ dotnet remove reference ../../src/FaceRecognitionDotNet/DlibDotNet.csproj
$ dotnet remove reference ../../src/FaceRecognitionDotNet/FaceRecognitionDotNet.csproj
$ dotnet add package FaceRecognitionDotNet
$ dotnet build -c Release
```

--------------------------------

### Train the Emotion Model

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/EmotionTraining/README.md

Initiate the emotion model training process using the .NET CLI. This command requires the dataset directory, epoch count, learning rates, batch size, validation interval, and output directory.

```bash
cd <EmotionTraining_dir>
dotnet run -c Release -- train --dataset <dataset_dir> \
                               --epoch 3000 \
                               --lr 0.0005 \
                               --min-lr 0.000000005 \
                               --min-batchsize 512 \
                               --validation-interval 30 \
                               --output <dataset_dir>\result
```

--------------------------------

### Visualize Training Log

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/GenderTraining/README.md

Generate a graph from the training log file to visualize the training process. This is useful for analyzing model performance over epochs.

```bash
python tools\visualize-log.py utkface-gender-network_600_0.001_1E-05_400.log
```

--------------------------------

### Preprocess Age Training Dataset

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/AgeTraining/README.md

Command to preprocess the created dataset by detecting faces, cropping face areas, and saving the cleaned data.

```bash
cd <AgeTraining_dir>
dotnet run -c Release -- preprocess --dataset=AdienceDataset --output=AdienceDataset_preprocessed
```

--------------------------------

### Generate XML File for Training

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/HelenTraining/README.md

Generates an XML file from model files, which is used for training. Specify the directory containing model files and the desired padding.

```bash
$ dotnet run -c Release -- generate -m "models" -p 50
```

--------------------------------

### Visualize Training Log

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/AgeTraining/README.md

Use the provided Python script to visualize the training log generated during the age model training process. This helps in analyzing the model's performance over epochs.

```python
python tools\visualize-log.py adience-age-network_600_0.001_1E-05_384_False.log
```

--------------------------------

### Configure Header File

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/AgeClassification/CMakeLists.txt

Configures a header file from a template, replacing placeholders.

```cmake
configure_file(
    ${PROJECT_SOURCE_DIR}/dlib/config.h.in
    ${PROJECT_SOURCE_DIR}/dlib/config.h
    @ONLY)

```

--------------------------------

### Create Dataset for Gender Training

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/GenderTraining/README.md

Use this PowerShell script to create a dataset for training and testing gender classification models. It splits images based on the TrainRate and specifies the output directory and maximum number of images.

```powershell
pwsh tools\CreateDataset.ps1 -InputDirectory UTKFace -TrainRate 8 -OutputDirectory UTKFaceDataset -Max 0
```

--------------------------------

### Add FaceRecognitionDotNet Package

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/FaceEncodingPerformance/README.md

Removes the local project reference and adds the FaceRecognitionDotNet NuGet package. Builds the project in Release configuration.

```bash
$ dotnet remove reference ../../src/FaceRecognitionDotNet\FaceRecognitionDotNet.csproj
$ dotnet add package FaceRecognitionDotNet
$ dotnet build -c Release
```

--------------------------------

### Create Dataset from 300W-LP

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/HeadPoseTraining/README.md

Use the CreateDataset.ps1 PowerShell script to generate training and testing datasets from the 300W-LP dataset. The TrainRate parameter controls the split ratio.

```powershell
pwsh tools\CreateDataset.ps1 -InputDirectory <300W-LP-Dataset_dir> ^
                             -TrainRate 8 ^
                             -OutputDirectory <300W-LP-YourDataset_dir> ^
                             -Max 0
```

--------------------------------

### Add FaceRecognitionDotNet Package

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/examples/CustomClassificationDemo/README.md

Removes the local project reference and adds the FaceRecognitionDotNet NuGet package to the project.

```bash
$ dotnet remove reference ../../src/FaceRecognitionDotNet\FaceRecognitionDotNet.csproj
$ dotnet add package FaceRecognitionDotNet
```

--------------------------------

### Initialize FaceRecognition with Model Directory

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Initializes the FaceRecognition instance by specifying the directory containing model files. It's recommended to set the internal encoding to 'shift_jis' to handle non-alphanumeric characters in paths.

```csharp
FaceRecognition.InternalEncoding = System.Text.Encoding.GetEncoding("shift_jis");

string directory = "C:\\models"
using (FaceRecognition fr = FaceRecognition.Create(directory))
```

--------------------------------

### Helen Face Landmark Detection

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Sets up a custom face landmark detector using the Helen dataset model. The 'helen-dataset.dat' file must be present in the 'models' directory.

```cs
using (var detector = new HelenFaceLandmarkDetector(Path.Combine("models", "helen-dataset.dat")))
{
    fr.CustomFaceLandmarkDetector= detector;

    var locations = fr.FaceLocations(image);
    var landmarks = fr.FaceLandmark(image, locations , PredictorModel.Custom);
}
```

--------------------------------

### Include Directories and Link Libraries

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/AgeClassification/CMakeLists.txt

Includes specified directories and links the dlib::dlib library to the project.

```cmake
include_directories(${INCLUDE_DIR})
target_link_libraries(${PROJ_NAME} dlib::dlib )

```

--------------------------------

### Run Gender Training

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/GenderTraining/README.md

Execute the gender training command with specified dataset, batch size, epochs, and validation interval. This command initiates the training process and outputs training progress and accuracy metrics.

```bash
cd <GenderTraining_dir>
dotnet run -c Release -- train -d=UTKFaceDataset -b=400 -e=600 -v=20
```

```text
              Epoch: 600
      Learning Rate: 0.001
  Min Learning Rate: 1E-05
     Min Batch Size: 400
Validation Interval: 20

Start load train images
Load train images: 18966
Start load test images
Load test images: 4738
step#: 0     learning rate: 0.001  average loss: 0            steps without apparent progress: 0
Epoch: 0, learning Rate: 0.001, average loss: 0.674069870253767
Epoch: 1, learning Rate: 0.001, average loss: 0.61758733607198
step#: 118   learning rate: 0.001  average loss: 0.59829      steps without apparent progress: 0
Epoch: 2, learning Rate: 0.001, average loss: 0.497301147655114
Epoch: 3, learning Rate: 0.001, average loss: 0.477524129911404
Epoch: 4, learning Rate: 0.001, average loss: 0.463913708851862
step#: 242   learning rate: 0.001  average loss: 0.463372     steps without apparent progress: 43
Epoch: 5, learning Rate: 0.001, average loss: 0.41626401307513
Epoch: 6, learning Rate: 0.001, average loss: 0.406071298713716
step#: 366   learning rate: 0.001  average loss: 0.400616     steps without apparent progress: 7
Epoch: 7, learning Rate: 0.001, average loss: 0.377037954321967
Epoch: 8, learning Rate: 0.001, average loss: 0.369290893870278
Epoch: 9, learning Rate: 0.001, average loss: 0.361628701590169
step#: 487   learning rate: 0.001  average loss: 0.361323     steps without apparent progress: 16
Epoch: 10, learning Rate: 0.001, average loss: 0.336463510539802
Saved state to utkface-gender-network_600_0.001_1E-05_400
Epoch: 11, learning Rate: 0.001, average loss: 0.332591397454274
step#: 606   learning rate: 0.001  average loss: 0.329499     steps without apparent progress: 43
Epoch: 12, learning Rate: 0.001, average loss: 0.309775507864488
Epoch: 13, learning Rate: 0.001, average loss: 0.306856178331174
Epoch: 14, learning Rate: 0.001, average loss: 0.302869322321159
step#: 726   learning rate: 0.001  average loss: 0.301861     steps without apparent progress: 28
Epoch: 15, learning Rate: 0.001, average loss: 0.29106002112316
Epoch: 16, learning Rate: 0.001, average loss: 0.285738143057821
step#: 850   learning rate: 0.001  average loss: 0.283248     steps without apparent progress: 72
Epoch: 17, learning Rate: 0.001, average loss: 0.271353094977893
Epoch: 18, learning Rate: 0.001, average loss: 0.270071162790127
Epoch: 19, learning Rate: 0.001, average loss: 0.265453205132239
step#: 974   learning rate: 0.001  average loss: 0.26501      steps without apparent progress: 43
Epoch: 20, learning Rate: 0.001, average loss: 0.253938037542761
Epoch: 20, train accuracy: 0.884266582305178, test accuracy: 0.878429717180245
...
...
Epoch: 590, learning Rate: 1E-05, average loss: 0.00314736557554334
step#: 28368  learning rate: 1e-05  average loss: 0.00314737   steps without apparent progress: 1422
Epoch: 591, Learning Rate: 1E-05, average loss: 0.00301913976009434
Saved state to utkface-gender-network_600_0.001_1E-05_400_
Epoch: 592, Learning Rate: 1E-05, average loss: 0.00311165843646179
step#: 28486  learning rate: 1e-05  average loss: 0.0030405    steps without apparent progress: 1543
Epoch: 593, Learning Rate: 1E-05, average loss: 0.00355719667702782
Epoch: 594, Learning Rate: 1E-05, average loss: 0.00323933126534789
step#: 28606  learning rate: 1e-05  average loss: 0.00313994   steps without apparent progress: 1664
Epoch: 595, Learning Rate: 1E-05, average loss: 0.00382428842364837
Epoch: 596, Learning Rate: 1E-05, average loss: 0.00292338314099893
Epoch: 597, Learning Rate: 1E-05, average loss: 0.00300246790072112
step#: 28727  learning rate: 1e-05  average loss: 0.00304047   steps without apparent progress: 1785
Epoch: 598, Learning Rate: 1E-05, average loss: 0.00299024601756618
Epoch: 599, Learning Rate: 1E-05, average loss: 0.00284008706881824
Saved state to utkface-gender-network_600_0.001_1E-05_400
done training
training num_right: 18928
training num_wrong: 38
training accuracy:  0.997996414636718
testing num_right: 4309
testing num_wrong: 429
testing accuracy:  0.909455466441537
```

--------------------------------

### Train Head Pose Model

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/HeadPoseTraining/README.md

Initiates the training process for a head pose model. Specify the dataset directory, training parameters like gamma and tolerance, and an output directory for results. The process involves loading and caching training and testing images for roll, pitch, and yaw, followed by training and validation steps.

```cmd
cd <HeadPoseTraining_dir>
dotnet run -c Release -- train --dataset <300W-LP-YourDataset_dir> ^
                      --gamma 0.1 ^
                      --tolerance 0.001 ^
                      --range 10 ^
                      --output <300W-LP-YourDataset_dir>\result

  Dataset: <300W-LP-YourDataset_dir>
Tolerance: 0.001
    Gamma: 0.1
    Range: 10
     Pose: All
   Output: <300W-LP-YourDataset_dir>\result

Start load train images
Use Cache <300W-LP-YourDataset_dir>\train\cache_Roll.dat
Load train images: 4000
Start load test images
Use Cache <300W-LP-YourDataset_dir>\test\cache_Roll.dat
Load test images: 1000

Training
100.00% Step 4000 of 4000                                  00:06:55
───────────────────────────────────────────────────────────────────
Validation for Training data
100.00% Step 4000 of 4000                                  00:00:01
───────────────────────────────────────────────────────────────────
training num_right: 4000
training num_wrong: 0
 training accuracy:  1.0000

Validation for Test data
100.00% Step 1000 of 1000                                  00:00:00
───────────────────────────────────────────────────────────────────
testing num_right: 872
testing num_wrong: 128
 testing accuracy:  0.8720
train accuracy: 1.0000, test accuracy: 0.8720

Start load train images
Use Cache <300W-LP-YourDataset_dir>\train\cache_Pitch.dat
Load train images: 4000
Start load test images
Use Cache <300W-LP-YourDataset_dir>\test\cache_Pitch.dat
Load test images: 1000

Training
100.00% Step 4000 of 4000                                  00:00:16
───────────────────────────────────────────────────────────────────
100.00% Step 4000 of 4000                                  00:00:00
───────────────────────────────────────────────────────────────────
training num_right: 3949
training num_wrong: 51
 training accuracy:  0.9873
100.00% Step 1000 of 1000                                  00:00:00
───────────────────────────────────────────────────────────────────
testing num_right: 969
testing num_wrong: 31
 testing accuracy:  0.9690
train accuracy: 0.9873, test accuracy: 0.9690

Start load train images
Use Cache <300W-LP-YourDataset_dir>\train\cache_Yaw.dat
Load train images: 4000
Start load test images
Use Cache <300W-LP-YourDataset_dir>\test\cache_Yaw.dat
Load test images: 1000
100.00% Step 4000 of 4000                                  00:00:27
───────────────────────────────────────────────────────────────────
100.00% Step 4000 of 4000                                  00:00:00
───────────────────────────────────────────────────────────────────
training num_right: 4000
training num_wrong: 0
 training accuracy:  1.0000
100.00% Step 1000 of 1000                                  00:00:00
───────────────────────────────────────────────────────────────────
testing num_right: 1000
testing num_wrong: 0
 testing accuracy:  1.0000
train accuracy: 1.0000, test accuracy: 1.0000
```

--------------------------------

### Enable NEON Support

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/AgeClassification/CMakeLists.txt

Configures C++ compiler flags for NEON support on ARM64 architecture.

```cmake
if (${ENABLE_NEON})
        # ARM64 ignore mfpu option
        # https://developer.arm.com/docs/dui0774/i/compiler-command-line-options/-mfpu
	#set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3 -mfpu=neon -fprofile-use -D ENABLE_NEON")
	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3 -fprofile-use -D ENABLE_NEON")
endif()

```

--------------------------------

### Configure and Build with CUDA Support

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/GenderClassification/CMakeLists.txt

Handles build configuration when CUDA is enabled. It disables static linking for CUDA runtime and appends the CUDA include path based on the operating system.

```cmake
if (${DLIB_USE_CUDA})
    configure_file(
        ${CMAKE_CURRENT_SOURCE_DIR}/version-cuda.rc.in
        ${CMAKE_CURRENT_BINARY_DIR}/version-cuda.rc
        @ONLY)

    # disable cuda static link
    set(CUDA_USE_STATIC_CUDA_RUNTIME OFF)

    # include CUDA Runtime Library
    if (MSVC OR MSYS OR MINGW)
        list(APPEND INCLUDE_DIR "$ENV{CUDA_PATH}/include")
    elseif(APPLE)
        list(APPEND INCLUDE_DIR "$ENV{CUDA_PATH}/include")
    elseif(UNIX AND NOT APPLE)
        list(APPEND INCLUDE_DIR "$ENV{CUDA_PATH}/include")
    else()
        message(FATAL_ERROR "Failed to link CUDA Runtime Library")
    endif()

    add_library(${PROJ_NAME} SHARED ${HEADERS}
                                    ${SOURCES}
                                    ${DLIBDOTNET_SHARED_HEADER1}
                                    ${DLIBDOTNET_SHARED_HEADER2}
                                    ${CMAKE_CURRENT_BINARY_DIR}/version-cuda.rc)
else()
    configure_file(
      ${CMAKE_CURRENT_SOURCE_DIR}/version.rc.in
      ${CMAKE_CURRENT_BINARY_DIR}/version.rc
      @ONLY)

    if ("${CMAKE_SYSTEM_NAME}" STREQUAL "iOS")
        add_library(${PROJ_NAME} ${HEADERS}
                                 ${SOURCES}
                                 ${DLIBDOTNET_SHARED_HEADER1}
                                 ${DLIBDOTNET_SHARED_HEADER2}
                                 ${CMAKE_CURRENT_BINARY_DIR}/version.rc)
    else()
        add_library(${PROJ_NAME} SHARED ${HEADERS}
                                        ${SOURCES}
                                        ${DLIBDOTNET_SHARED_HEADER1}
                                        ${DLIBDOTNET_SHARED_HEADER2}
                                        ${CMAKE_CURRENT_BINARY_DIR}/version.rc)
    endif()
endif()

```

--------------------------------

### Train Helen Dataset

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/HelenTraining/README.md

Trains the model using the generated XML file and specifies the number of threads to use.

```bash
$ dotnet run -c Release -- train -x "helen\helen-dataset.xml" -t 8
```

--------------------------------

### Set Include Directory

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/AgeClassification/CMakeLists.txt

Sets the include directory for the project.

```cmake
set(INCLUDE_DIR "${PROJECT_SOURCE_DIR}/../DlibDotNet/src")

```

--------------------------------

### Initialize FaceRecognition with Model Binary Data

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/wiki/Quickstart

Initializes the FaceRecognition instance by providing model binary data directly. This method is useful when model files are embedded as resources.

```csharp
var modelParameter = new ModelParameter
{
    PosePredictor68FaceLandmarksModel = File.ReadAllBytes("shape_predictor_68_face_landmarks.dat"),
    PosePredictor5FaceLandmarksModel = File.ReadAllBytes("shape_predictor_5_face_landmarks.dat"),
    FaceRecognitionModel = File.ReadAllBytes("dlib_face_recognition_resnet_model_v1.dat"),
    CnnFaceDetectorModel = File.ReadAllBytes("mmod_human_face_detector.dat")
}
using (FaceRecognition fr = FaceRecognition.Create(modelParameter ))
```

--------------------------------

### CPack Configuration

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/AgeClassification/CMakeLists.txt

Sets project name and version for CPack, the CMake packaging utility.

```cmake
set(CPACK_PROJECT_NAME ${PROJECT_NAME})
set(CPACK_PROJECT_VERSION ${PROJECT_VERSION})
include(CPack)

```

--------------------------------

### Configure CUDA Build Settings

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/EmotionClassification/CMakeLists.txt

Sets up build configurations when DLIB_USE_CUDA is enabled. It disables static linking of the CUDA runtime and appends the CUDA include directory.

```cmake
if (${DLIB_USE_CUDA})
	configure_file(
	    ${CMAKE_CURRENT_SOURCE_DIR}/version-cuda.rc.in
	    ${CMAKE_CURRENT_BINARY_DIR}/version-cuda.rc
	    @ONLY)

	set(CUDA_USE_STATIC_CUDA_RUNTIME OFF)

	if (MSVC OR MSYS OR MINGW)
	    list(APPEND INCLUDE_DIR "$ENV{CUDA_PATH}/include")
	elseif(APPLE)
	    list(APPEND INCLUDE_DIR "$ENV{CUDA_PATH}/include")
	elseif(UNIX AND NOT APPLE)
	    list(APPEND INCLUDE_DIR "$ENV{CUDA_PATH}/include")
	else()
	    message(FATAL_ERROR "Failed to link CUDA Runtime Library")
	endif()

```

--------------------------------

### Enable ARM NEON Optimization

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/src/GenderClassification/CMakeLists.txt

Applies optimization flags for ARM64 architecture when NEON is enabled. It sets optimization level to O3 and enables profile-guided optimization.

```cmake
if (${ENABLE_NEON})
        # ARM64 ignore mfpu option
        # https://developer.arm.com/docs/dui0774/i/compiler-command-line-options/-mfpu
	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3 -fprofile-use -D ENABLE_NEON")
endif()

```

--------------------------------

### Run Emotion Recognition Model Test

Source: https://github.com/takuya-takeuchi/facerecognitiondotnet/blob/master/tools/EmotionTrainingV2/README.md

Execute the test command for emotion recognition models. Ensure the 'Models' directory is prepared and specify the dataset and model paths. This command initiates the testing process and outputs detailed logs.

```cmd
cd <EmotionTraining_dir>
dotnet run -c Debug -- test --dataset <dataset_dir> ^
                            --model "Corrective_re-annotation_of_FER_CK+_KDEF-mlp_3000_5E-05_5E-09_512.dat"
```
Tokens:
Raw