为什么我在运行此脚本时收到 tensorflow 警告？

Question

import glob
import os
from mtcnn.mtcnn import MTCNN
import warnings
import time

from numpy import asarray
from PIL import Image
#warnings.filterwarnings("ignore")
#os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

directory = input("insert input path \n")

output_directory = input("insert output path \n")
#mode=input("do you want to conver the outputs to Grayscale ?")
img_names=glob.glob(os.path.join(directory+"/*.jpg"))

detector = MTCNN()
def process_image(img_name,mode='L',output_size=(160,160)):
    img = Image.open(directory+img_name)
    img.thumbnail((160,160))
    pixels=asarray(img)
    results = detector.detect_faces(pixels)
    if results:
        # extract the bounding box from the requested face
        x1 ,y1,width,height=results[0]['box']
        x1,y1=abs(x1),abs(y1)
        x2,y2=x1 + width,y1 + height
        # extract the face by slicing
        face_boundary = pixels[y1:y2, x1:x2]
        # resize pixels to the model size
        #image1 = Image.fromarray(face_boundary)
        #image1 = image.resize(required_size)
        image=Image.fromarray(face_boundary)
        #if mode=='L':
         #   image=image.convert('L')
        image = image.resize(output_size)
        #image.thumbnail((160,160))
        #image = image.resize(())
        #face_array = asarray(image)
    #image.save(f"/kaggle/input/rashaa/rasha{img_name}")
        image.save(f'{output_directory}{img_name}')     
        print(f'{img_name} was processed...')
#for img in img_names:
 #       x.append(img.replace(directory,""))
x=[img.replace(directory,"") for img in img_names]
t1 = time.perf_counter()
y=[process_image(img) for img in x]

t2=time.perf_counter()
print(t2-t1)

the code does the job by detecting and extracting the faces from the input folder and putting the extracted faces in the output folder without any issues but i wanna know why is this warning is showing up in the first place and is there any way i can fix it "properly" instead of suppressing it

details

• TensorFlow version (CPU):2.7.0

• python version 3.8.4

the warning message is WARNING:tensorflow:5 out of the last 9 calls to <function Model.make_predict_function..predict_function at 0x0000000013E161F0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer tohttps://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.

Answer 1

In your case warning happens because of (2) passing tensors with different shapes . And you get tensors of different shape because of PIL. img.thumbnail resizes some of your images in a wrong way. It keeps the aspect ratio of the image, so you get 160x160 image only if your source image had 1:1 aspect ratio.

Use Tensorflow or OpenCV to process images, not PIL it's very slow...

def process_image(img_name, output_size=(160,160)):
    img = cv2.imread(directory+img_name)
    img = img[..., ::-1] # convert BGR to RGB
    img = cv2.resize(img, (160, 160))
    results = detector.detect_faces(img)
    ...