对图片/视频中的人脸进行检测，并绘制人脸框。然后对检测到的人脸进行关键点识别，并进行绘制。最后根据人脸关键点，裁剪出人脸，判断该人脸的表情。

基于此，分别使用retina进行人脸检测，PFLD进行人脸关键点识别，emotion-ferplus-8进行人脸表情的分类，它是基于微软的 FER+ 数据集训练的CNN分类网络，准确率约62%。

模型地址在文末...

代码如下，有详细的注释，注意修改一下自己的模型地址。

import sys
import os
import cv2
import numpy as np
import argparse
import logging.config
from torchvision import transforms


# 添加路径一次即可
sys.path.append('/data/ai/FaceSDK')

from core.image_cropper.arcface_cropper.FaceRecImageCropper import FaceRecImageCropper
from models.model_pipline import ModelLoader

# 配置日志
mpl_logger = logging.getLogger('matplotlib')
mpl_logger.setLevel(logging.WARNING)
logging.config.fileConfig("/data/ai/FaceSDK/config/logging.conf")
logger = logging.getLogger('api')

# 初始化模型加载器，加载模型
model_path = '/data/ai/FaceSDK/models'
model_loader = ModelLoader(model_path)

# 获取需要的模型处理器
faceDetModelHandler = model_loader.get_face_det_model_handler()  # 人脸检测
faceAlignModelHandler = model_loader.get_face_align_model_handler()  # 人脸对齐
face_cropper = FaceRecImageCropper()  # 人脸裁剪

emo_model = cv2.dnn.readNetFromONNX('/data/ai/FaceSDK/models/emotion-ferplus-8.onnx')  # 加载onnx人脸表情模型
logger.info(f"人脸表情识别模型加载成功....")

# 定义情感字典
emotion_dict = {
    0: 'neutral',
    1: 'happiness',
    2: 'surprise',
    3: 'sadness',
    4: 'anger',
    5: 'disgust',
    6: 'fear'
}


def emotion_process(image_name, image):
    dets = faceDetModelHandler.inference_on_image(image)  # 人脸检测
    face_nums = dets.shape[0]

    bboxs = dets
    for i in range(face_nums):
        box = list(map(int, bboxs[i]))
        cv2.rectangle(image, (box[0], box[1]), (box[2], box[3]), (0, 0, 255), 2)  # 绘制人脸检测框

        landmarks = faceAlignModelHandler.inference_on_image(image, bboxs[i])
        for (x, y) in landmarks.astype(np.int32):
            cv2.circle(image, (x, y), 2, (255, 0, 0), -1)  # 绘制人脸关键点

        landmarks_list = []
        for (x, y) in landmarks.astype(np.int32):
            landmarks_list.extend((x, y))
        cropped_image = face_cropper.crop_image_by_mat(image, landmarks_list)  # 裁剪人脸

        # 调整尺寸并转换为单通道灰度图像
        cropped_image = cv2.resize(cropped_image, (64, 64))
        cropped_image = cv2.cvtColor(cropped_image, cv2.COLOR_RGB2GRAY)  # 转换为灰度图像
        blob = cv2.dnn.blobFromImage(cropped_image, scalefactor=1.0, size=(64, 64), mean=(0, 0, 0), swapRB=False,
                                     crop=False)

        # 确保输入张量的形状正确
        if blob.shape[1:] != (1, 64, 64):
            logger.error(f"Incorrect blob shape: {blob.shape[1:]}. Expected (1, 64, 64).")
            continue

        emo_model.setInput(blob)
        try:
            output = emo_model.forward()
        except Exception as e:
            logger.error(f"Error during model inference: {e}")
            continue
        pred = emotion_dict[list(output[0]).index(max(output[0]))]  # 推理人脸表情

        # 绘制情感标签
        cv2.putText(
            image,
            pred,
            (box[0], box[1] - 10),
            cv2.FONT_HERSHEY_SIMPLEX,
            0.8,
            (215, 5, 247),
            2,
            lineType=cv2.LINE_AA
        )

    return image


def process_image(image_path, result_folder):
    image_name = os.path.basename(image_path)
    try:
        image = cv2.imread(image_path, cv2.IMREAD_COLOR)
        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
        processed_image = emotion_process(image_name, image)
        result_path = os.path.join(result_folder, image_name)
        cv2.imwrite(result_path, cv2.cvtColor(processed_image, cv2.COLOR_RGB2BGR))
        logger.info(f"Result saved for image: {image_name}")
    except Exception as e:
        logger.error(f"Error processing image {image_path}: {e}")


def process_video(video_path, result_folder):
    cap = cv2.VideoCapture(video_path)
    if not cap.isOpened():
        logger.error(f"Error opening video file {video_path}")
        return

    frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    size = (frame_width, frame_height)
    result_path = os.path.join(result_folder, 'result_video.avi')
    result = cv2.VideoWriter(result_path, cv2.VideoWriter_fourcc(*'MJPG'), 10, size)

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break

        frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        processed_frame = emotion_process('frame', frame_rgb)
        result.write(cv2.cvtColor(processed_frame, cv2.COLOR_RGB2BGR))

    cap.release()
    result.release()
    logger.info(f"Result video saved to {result_path}")


def process_folder(folder_path, result_folder):
    for image_name in os.listdir(folder_path):
        image_path = os.path.join(folder_path, image_name)
        process_image(image_path, result_folder)


def main(args):
    # 检查并创建结果文件夹
    if not os.path.exists(args.result_folder):
        os.makedirs(args.result_folder)

    if os.path.isdir(args.src):
        process_folder(args.src, args.result_folder)
    elif args.src.lower().endswith(('.mp4', '.avi', '.mov')):
        process_video(args.src, args.result_folder)
    else:
        process_image(args.src, args.result_folder)


if __name__ == '__main__':
    parser = argparse.ArgumentParser(description="Detect Face Emotion")
    parser.add_argument('--src', type=str, required=True, help='path to detect data (image, video, or folder)')
    parser.add_argument('--result_folder', type=str, required=True, help='path to save results')
    args = parser.parse_args()
    main(args)

运行脚本

python emotion_pipline.py --src /data/ai/FaceSDK/emotion/test\
                --result_folder /data/ai/FaceSDK/emotion/result

仓库地址：GitHub - JDAI-CV/FaceX-Zoo: A PyTorch Toolbox for Face Recognition