TPRD-FER-Official-Code-Implementations

Official Implementations of "Text Prompt Region Decomposition for Effective Facial Expression Recognition"

Our manuscript is currently under review; therefore, this project only provides inference code and weight files for testing.

The training code will be availabel later.

---

Installation

1. Installation the package requirements

pip install -r requirements.txt

---

Data Preparation

1. The reorganized version of RAF-DB can be downloaded at APViT:

data/
├─ RAF-DB/
│  ├─ basic/
│  │  ├─ EmoLabel/
│  │  │  ├─ list_patition_label.txt
│  │  │  ├─ rafdb_occlusion_list.txt
│  │  │  ├─ val_raf_db_list.txt
│  │  │  ├─ val_raf_db_list_45.txt
│  │  ├─ Image/
│  │  │  ├─ aligned_224/  # reagliend by MTCNN

2. The downloaded AffectNet are organized as follow:

data/
├─ AffectNet/
│  ├─ Manually_Annotated_Images/
│  │  ├─ training.csv
│  │  ├─ validation.csv
│  │  ├─ 1/
│  │  │  ├─ images
│  │  │  ├─ ...
│  │  ├─ 2/
│  │  ├─ ./

3. The FER2013, FERPlus, and pre-processing code are available at https://github.com/microsoft/FERPlus:

4. The Occlusion- and Pose- variant lists can be downloaded at RAN:

---

Model checkpoints

• Download model checkpoints from Google Drive.
• Modify the path of the downloaded weight files in the configuration files.

---

Testing

python test.py --config ${CONFIG_PATH}