Data: RAF-DB
GCN framework: GCNconv
Ref : MRE-CNN
Mediapipe model: face_landmarker
Frontend template: free css/html template
- Download the dataset from Kaggle and place it in the Image_data folder.
- Install the face_landmarker tool and place it in the model folder.
- Execute
pip install -r requirements.txtfor packages. - Execute
RunAdjacency.pyto generate the image adjacency matrix. - Execute
Fileprocess.pyto preprocess the data for each category. - Execute
RunModel.pyto train the model with mlflow.
- Execute
Download.pyto get the models for frontend, the models including Mediapipe and the models generated byRunModel.py - Execute
app.pyscript to run the web application.
UMAP.py provides a lower-dimensional view of the model.
path/GCNN
├─.ipynb_checkpoints
├─GradCam
├─Image_data
│ └─DATASET
│ ├─test
│ │ ├─1
│ │ ├─2
│ │ ├─3
│ │ ├─4
│ │ ├─5
│ │ ├─6
│ │ └─7
│ └─train
│ ├─1
│ ├─2
│ ├─3
│ ├─4
│ ├─5
│ ├─6
│ └─7
├─model
├─output_data
│ ├─adjacency
│ │ ├─adjacency_1
│ │ ├─adjacency_2
│ │ ├─adjacency_3
│ │ ├─adjacency_4
│ │ ├─adjacency_5
│ │ ├─adjacency_6
│ │ └─adjacency_7
│ └─landmarks
Install torch with the command below to get a GPU version (if any)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/metal.html
尹士文 | 周雨舒 | 張藝馨 | 蕭邦宇 |