app/image_processing.py

import numpy as np
import cv2
from PIL import Image
import torch
from app.model import pth_model_static, cam, pth_processing
from app.face_utils import get_box
from app.config import DICT_EMO
from pytorch_grad_cam.utils.image import show_cam_on_image
import mediapipe as mp

mp_face_mesh = mp.solutions.face_mesh

def preprocess_image_and_predict(inp):
    inp = np.array(inp)

    if inp is None:
        return None, None, None

    try:
        h, w = inp.shape[:2]
    except Exception:
        return None, None, None

    with mp_face_mesh.FaceMesh(
        max_num_faces=1,
        refine_landmarks=False,
        min_detection_confidence=0.5,
        min_tracking_confidence=0.5,
    ) as face_mesh:
        results = face_mesh.process(inp)
        if results.multi_face_landmarks:
            for fl in results.multi_face_landmarks:
                startX, startY, endX, endY = get_box(fl, w, h)
                cur_face = inp[startY:endY, startX:endX]
                cur_face_n = pth_processing(Image.fromarray(cur_face))
                with torch.no_grad():
                    prediction = (
                        torch.nn.functional.softmax(pth_model_static(cur_face_n), dim=1)
                        .detach()
                        .numpy()[0]
                    )
                confidences = {DICT_EMO[i]: float(prediction[i]) for i in range(7)}
                grayscale_cam = cam(input_tensor=cur_face_n)
                grayscale_cam = grayscale_cam[0, :]
                cur_face_hm = cv2.resize(cur_face,(224,224))
                cur_face_hm = np.float32(cur_face_hm) / 255
                heatmap = show_cam_on_image(cur_face_hm, grayscale_cam, use_rgb=True)

    return cur_face, heatmap, confidences