app.py

import gradio as gr
import torch
from PIL import Image
# from PIL import ImageDraw
# from PIL import ImageFont
import cv2
import numpy as np
import argparse

def parse_args(known=False):
  parser = argparse.ArgumentParser(description="Object Counting v1.0")
  parser.add_argument(
      "--yolov5_path",
      "-yp",
      default="ultralytics/yolov5",
      type=str,
      help="yolov5 path",
  )
  parser.add_argument(
      "--model_path",
      "-mp",
      default="model/yolov5n_rebar_kaggle.pt",
      type=str,
      help="model path",
  )
  parser.add_argument(
      "--nms_conf",
      "-conf",
      default=0.25,
      type=float,
      help="model NMS confidence threshold",
  )
  parser.add_argument(
      "--nms_iou", 
      "-iou", 
      default=0.1, 
      type=float, 
      help="model NMS IoU threshold"
  )
  parser.add_argument(
      "--device",
      "-dev",
      default="cpu",
      type=str,
      help="cuda or cpu",
  )
  parser.add_argument("--inference_size", "-isz", default=640, type=int, help="model inference size")
  parser.add_argument("--slider_step", "-ss", default=0.05, type=float, help="slider step")

  args = parser.parse_known_args()[0] if known else parser.parse_args()
  return args

def image_roi(im):
  mask = np.array(im["mask"])
  mask = cv2.cvtColor(mask, cv2.COLOR_BGR2GRAY)
  contours, hierarchy = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
  if contours:
    mask = np.zeros(mask.shape, np.uint8)
    cnt = contours[0]
    mask = cv2.drawContours(mask, [cnt], 0, 255, -1)
    im = np.array(im["image"])
    im = cv2.bitwise_and(im, im, mask=mask)
    im = Image.fromarray(im)
  else:
    im = im["image"] 
  return im

def yolo(im, conf, iou, size):

  global model
  if im is None:
    print ("No image")
    return None, None
  im = image_roi(im)
  model.conf = conf
  model.iou = iou
  results = model(im, size=size)  # custom inference size
  output_im = np.array(im)

  pred = results.pandas().xyxy[0]
  counting = pred.shape[0]
  text = f"{counting} objects"
  for index, row in pred.iterrows():
    cv2.circle(output_im, (int((row["xmin"] + row["xmax"]) * 0.5), int((row["ymin"] + row["ymax"]) * 0.5)), int((row["xmax"] - row["xmin"]) * 0.5 * 0.6), (255, 0, 0), -1)

  return Image.fromarray(output_im), text


def main(args):
  gr.close_all()

  global model
  yolo_path = args.yolov5_path
  model_path = args.model_path
  nms_conf = args.nms_conf
  nms_iou = args.nms_iou
  device = args.device
  inference_size = args.inference_size
  slider_step = args.slider_step

  model = torch.hub.load(yolo_path, 'custom', path=model_path, device=device)
  inputs_image = gr.inputs.Image(tool="sketch", label="Original Image",type="pil")
  inputs_conf = gr.Slider(0, 1, step=slider_step, value=nms_conf, label="Conf Thres")
  inputs_iou = gr.Slider(0, 1, step=slider_step, value=nms_iou, label="IoU Thres")
  inputs_size = gr.Slider(384, 1536, step=128, value=inference_size, label="Inference Size")
  inputs = [inputs_image, inputs_conf, inputs_iou, inputs_size]

  outputs_image = gr.outputs.Image(type="pil", label="Output Image")
  outputs_text = gr.Textbox(label="Number of objects")
  outputs = [outputs_image, outputs_text]

  title = "OBJECT COUNTING DEMO"
  description = "Upload an image or click an example image to use. You can select the area to count by drawing a closed area on the input image."
  article = "<p style='text-align: center'>Counting objects in image</a></p>"
  examples = [['./images/S__275668998.jpg'], ['./images/S__275669003.jpg'], ['./images/S__275669004.jpg']]

  gr.Interface(fn=yolo, inputs=inputs, outputs=outputs, title=title, description=description, article=article, examples=examples, cache_examples=False, analytics_enabled=False).launch(
      debug=True)#, share=True)


if __name__ == "__main__":
    args = parse_args()
    main(args)