app.py

#!/usr/bin/env python

from __future__ import annotations

import os
import pathlib
import sys

import cv2
import gradio as gr
import numpy as np
import torch

sys.path.insert(0, "face_detection")
sys.path.insert(0, "face_alignment")
sys.path.insert(0, "emotion_recognition")

from ibug.emotion_recognition import EmoNetPredictor
from ibug.face_alignment import FANPredictor
from ibug.face_detection import RetinaFacePredictor

DESCRIPTION = "# [ibug-group/emotion_recognition](https://github.com/ibug-group/emotion_recognition)"

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

face_detector = RetinaFacePredictor(threshold=0.8, device=device, model=RetinaFacePredictor.get_model("mobilenet0.25"))
landmark_detector = FANPredictor(device=device, model=FANPredictor.get_model("2dfan2"))

model_names = [
    "emonet248",
    "emonet245",
    "emonet248_alt",
    "emonet245_alt",
]
models = {name: EmoNetPredictor(device=device, model=EmoNetPredictor.get_model(name)) for name in model_names}


def predict(image: np.ndarray, model_name: str, max_num_faces: int) -> np.ndarray:
    model = models[model_name]
    if len(model.config.emotion_labels) == 8:
        colors: tuple[tuple[int, int, int], ...] = (
            (192, 192, 192),
            (0, 255, 0),
            (255, 0, 0),
            (0, 255, 255),
            (0, 128, 255),
            (255, 0, 128),
            (0, 0, 255),
            (128, 255, 0),
        )
    else:
        colors = (
            (192, 192, 192),
            (0, 255, 0),
            (255, 0, 0),
            (0, 255, 255),
            (0, 0, 255),
        )

    # RGB -> BGR
    image = image[:, :, ::-1]

    faces = face_detector(image, rgb=False)
    if len(faces) == 0:
        raise gr.Error("No face was found.")
    faces = sorted(list(faces), key=lambda x: -x[4])[:max_num_faces]
    faces = np.asarray(faces)
    _, _, features = landmark_detector(image, faces, rgb=False, return_features=True)
    emotions = model(features)

    res = image.copy()
    for index, face in enumerate(faces):
        box = np.round(face[:4]).astype(int)
        cv2.rectangle(res, tuple(box[:2]), tuple(box[2:]), (0, 255, 0), 2)

        emotion = emotions["emotion"][index]
        valence = emotions["valence"][index]
        arousal = emotions["arousal"][index]
        emotion_label = model.config.emotion_labels[emotion].title()

        text_content = f"{emotion_label} ({valence: .01f}, {arousal: .01f})"
        cv2.putText(
            res, text_content, (box[0], box[1] - 10), cv2.FONT_HERSHEY_DUPLEX, 1, colors[emotion], lineType=cv2.LINE_AA
        )

    return res[:, :, ::-1]


with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Row():
        with gr.Column():
            image = gr.Image(label="Input", type="numpy")
            model_name = gr.Radio(
                label="Model",
                choices=model_names,
                value=model_names[0],
                type="value",
            )
            max_num_of_faces = gr.Slider(
                label="Max Number of Faces",
                minimum=1,
                maximum=30,
                step=1,
                value=30,
            )
            run_button = gr.Button()
        with gr.Column():
            result = gr.Image(label="Output")
    gr.Examples(
        examples=[[path.as_posix(), model_names[0], 30] for path in sorted(pathlib.Path("images").rglob("*.jpg"))],
        inputs=[image, model_name, max_num_of_faces],
        outputs=result,
        fn=predict,
        cache_examples=os.getenv("CACHE_EXAMPLES") == "1",
    )
    run_button.click(
        fn=predict,
        inputs=[image, model_name, max_num_of_faces],
        outputs=result,
        api_name="predict",
    )

if __name__ == "__main__":
    demo.queue(max_size=20).launch()