app.py

from keras.models import load_model
from PIL import Image, ImageOps
import numpy as np
import gradio as gr
import pandas as pd
import torch
import json

# Load the model
classify_model = load_model('keras_model.h5')
detect_model = torch.hub.load('ultralytics/yolov5', 'custom', path='best.pt', force_reload=True, _verbose=False)

def detect(image):
  # Inference
  results = detect_model(image)
  try:
    results = json.loads(results.pandas().xyxy[0].to_json(orient="records"))[0]
    top = int(results['ymin'])
    left = int(results['xmin'])
    width = int(results['xmax'] - results['xmin'])
    height = int(results['ymax'] - results['ymin'])
    return top, left, width, height
  except:
    return 0,0,0,0

def format_label(label):
  """
  From '0 class 1\n' to 'class 1'
  """
  return label[:-1]


def predict(image):
  top, left, width, height = detect(image)

  if (top == 0) and (left == 0) and (width == 0) and (height==0):
    return {
      "predictions": {}, 
      'bbox': {
          "top": 0,
          "left": 0,
          "width": 0,
          "height": 0
      }
    }

  if width > height:
    height = width
  else:
    width = height
  
  # Crop the turtle
  image = image.crop((left, top, left + width, top + height))

  # Create the array of the right shape to feed into the keras model
  # The 'length' or number of images you can put into the array is
  # determined by the first position in the shape tuple, in this case 1.
  data = np.ndarray(shape=(1, 224, 224, 3), dtype=np.float32)
  
  #resize the image to a 224x224 with the same strategy as in TM2:
  #resizing the image to be at least 224x224 and then cropping from the center
  size = (224, 224)
  image = ImageOps.fit(image, size, Image.ANTIALIAS)

  #turn the image into a numpy array
  image_array = np.asarray(image)
  # Normalize the image
  normalized_image_array = (image_array.astype(np.float32) / 127.0) - 1
  # Load the image into the array
  data[0] = normalized_image_array

  # run the inference
  pred = classify_model.predict(data)
  pred = pred.tolist()

  with open('labels.txt','r') as f:
    labels = f.readlines()
    
  result = {format_label(labels[i]): round(pred[0][i],2) for i in range(len(pred[0]))}
  sorted_result = {k: v for k, v in sorted(result.items(), key=lambda item: item[1], reverse=True) if v > 0}

  
  return json.dumps({
      "predictions": sorted_result, 
      'bbox': {
          "top": top,
          "left": left,
          "width": width,
          "height": height
      }
    })
    
title = "🐆"

gr.Interface(
    fn=predict, 
    inputs=gr.Image(type="pil", label="Input Image"),
    outputs=[gr.JSON()],
  #  live=True,
    title=title,
).launch(share=True, debug=False)