water_meter_app.py

import pandas as pd
import numpy as np
import os, sys
# import seaborn as sns
import matplotlib.pyplot as plt
import glob
# from PIL import Image, ImageOps
# from plotly.subplots import make_subplots
# from warnings import filterwarnings
from io import StringIO, BytesIO
import streamlit as st
import torch
import cv2
# import argparse
from tempfile import NamedTemporaryFile
import warnings
from yolov5 import detect_yolov5
from PIL import Image


# ap = argparse.ArgumentParser()
# ap.add_argument("-i", "--img", required=False, help="path to image")
# args = vars(ap.parse_args())

def load_image(img_path, resize=True):
  img = cv2.cvtColor(cv2.imread(str(img_path)), cv2.COLOR_BGR2RGB)
  img = cv2.resize(img, (300, 300), interpolation = cv2.INTER_AREA)
  # img = Image.open(img_path)
  # img = ImageOps.exif_transpose(img.resize((300, 300), Image.ANTIALIAS))
  return img

def show_grid(image_paths):
  images = [load_image(img) for img in image_paths]
  images = torch.as_tensor(images)
  images = images.permute(0, 3, 1, 2)
  # grid_img = torchvision.utils.make_grid(images, nrow=11)
  plt.figure(figsize=(24, 12))
  # plt.imshow(grid_img.permute(1, 2, 0))
  plt.axis('off')

def format_predictions(img_path, results, num_classes=8):
  """
  Format the predictions as they are not in order of the meter reading
  For some reason inside this app, the  detect script generates more rows than
  it does when tested in Colab. So, I have taken the top num_classes records
  """
  if not isinstance(results, str):
    df = results.pandas().xyxy[0].head(num_classes)
  else:
    df = pd.read_csv(results, sep=' ', header=None)
  df.columns = ['xmin', 'ymin', 'xmax', 'ymax', 'conf', 'class', 'name']
  df.sort_values('xmin', ascending=True, inplace=True)
  img = load_image(img_path)

  fig = plt.figure(figsize=(7, 7))
  plt.imshow(img)
  print('df.values',df['class'].values)
  reading = ''
  for s in df['class'].values:
    # 防止把仪表识别结果误加入读数中
    if s != 10:
        reading = reading + str(s)
  # reading = ''.join(str(s) for s in df['class'].values)
  # if len(reading) > 5:
  #   reading = reading[:-3] + '.' + reading[-3:]
  # reading = reading[2:]
  try:
    reading = float(reading)
#     plt.title(str(img_path) + "\n" + str(img) + "\n" + "Predict labels: " + str(df['class'].values) + "\n" + 'Reading: ' +  str(reading) + " m\u00b3")
    plt.title("Predict labels: " + str(df['class'].values) + "\n" + 'Reading: ' +  str(reading) + " m\u00b3")
  except:
    plt.title('Is this a valid water meter image...?')
  plt.axis('off')
  st.pyplot(fig)

@st.cache
def load_model(src, path, device, reload=False):
  return torch.hub.load(src, 'custom', path=path, device=device, force_reload=reload)

@st.cache
def detect(weights, source):
#   os.system("python yolov5-master/detect.py --weights %s --source %s" % (weights, source))
#   img = load_image(source)
  im0 = detect_yolov5.run(weights, source)
  fig = plt.figure(figsize=(7, 7))
  plt.imshow(im0)
  plt.axis('off')
  st.pyplot(fig)

@st.cache(ttl=24*3600, suppress_st_warning=True, show_spinner=False)
def predict(inp):
# results = model(args['img'])
# format_predictions(args['img'], results)
  if inp:
    # https://discuss.streamlit.io/t/image-upload-problem/4810/5
    temp_file = NamedTemporaryFile(delete=True)
    temp_file.write(inp.getvalue())
    col1, col2 = st.columns(2)
    with col1:
      st.header("Original picture")
      results = model.to(device)(temp_file.name)
      format_predictions(temp_file.name, results)
    with col2:
#       st.header("Predict picture")
      st.header("Predict picture")
#       results = model.to(device)(temp_file.name)
#       format_predictions(temp_file.name, results)
      weights = 'weights/best.pt'
#       detect(weights, temp_file.name)
      img = Image.open(inp)
      detect(weights, inp)

if __name__ == '__main__':
  yolo_path = os.path.join(os.getcwd(), 'yolov5')
  # location = 'runs/train/yolov5x_water_meter/weights/best.pt'
  location = 'weights/best.pt'
  best_run = os.path.join(os.getcwd(), 'yolov5', location)
  device = torch.cuda.get_device_properties(0).name if torch.cuda.is_available() else 'cpu'

  # model = torch.hub.load('ultralytics/yolov5', 'custom', path=location, device=device, force_reload=False)
  src = 'ultralytics/yolov5'
  model = load_model(src, path=location, device=device)
  warnings.filterwarnings('ignore')
  st.title('Water meter reading system based on YOLOv5')
  buffer = st.file_uploader("Upload image", type=['png', 'jpeg', 'jpg'])
  predict(buffer)