import gradio as gr
import torch
import os
import requests
import json
import cv2
from PIL import Image
from timeit import default_timer as timer
import numpy as np
import aiofiles
from transformers import AutoModel

model = torch.hub.load('ultralytics/yolov5','yolov5s', pretrained=True)
#model1 = AutoModel.from_pretrained(model)
cnt = 0
def LCR(bbox,x_img, y_img):
    x1 = bbox[0]/x_img
    x2 = bbox[2]/x_img
    if x1 < 0.2 and x2 < 0.2 :
        location = "Left"
    elif x1 > 0.8 and x2 > 0.8:
        location = "Right"
    elif x1 < 0.2 and (x2 <= 0.8 and x2 >= 0.2):
        if (x1 + x2) < 0.4:
            location = "Left"
        else:
            location = "Center" 
    elif x2 > 0.8 and (x1 <= 0.8 and x1 >= 0.2):
        if (x1 + x2) > 1.6:
            location = "Right"
        else:
            location = "Center" 
    else:
        location = "Center"
    print(f"x1 {x1} x2 {x2} bbox0 {bbox[0]} bbox2 {bbox[2]} x_img {x_img} LocationLCR {location}")
    return location

def ACB(bbox, x_img, y_img, location):
    y1 = bbox[1]/y_img
    y2 = bbox[3]/y_img
    if location == "Center":
        if y1 < 0.33333 and y2 < 0.33333 :
            location = "Above"
        elif y1 > 0.66667 and y2 > 0.66667:
            location = "Below"
        elif y1 < 0.33333 and (y2 <= 0.66667 and y2 >= 0.33333):
            if (y1 + y2) < 0.66667:
                location = "Above"
            else:
                location = "Center" 
        elif y2 > 0.66667 and (y1 <= 0.66667 and y1 >= 0.33333):
            if (y1 + y2) > 1.33333:
                location = "Below"
            else:
                location = "Center" 
        else:
            location = "Center"
    else:
        pass
    print(f"y1 {y1} y2 {y2} bbox1 {bbox[1]} bbox3 {bbox[3]} y_img {y_img} Location{location}")
    
    return location
    #print(bbox[0])
    
def turn_img_into_fileJSON(frame):
    start_time = timer()
    x_img, y_img = frame.size
    print(x_img,y_img)
    global cnt
    objects = []

    prediction = model(frame)
    for det in prediction.xyxy[0]:
        class_id = int(det[5])
        class_name = model.names[class_id]
        confidence = float(det[4])
        bbox = det[:4].tolist()
        if(confidence >= 0.5):
            location = LCR(bbox, x_img, y_img)
            location = ACB(bbox, x_img, y_img, location)
            # Save the results to the list
            objects.append({
                'Class': class_name,
                #'BoundingBox': bbox,
                'Location': location,
                'Confidence': confidence
            })
        with open('{:05d}.json'.format(cnt)  , 'w') as f:
            json.dump(objects, f)
    cnt += 1
    pred_time = round(timer() - start_time, 5)
    json_str = json.dumps(objects)
    return json_str, pred_time
#path = [["D:/cuoc_thi/object-detection/download.jpg"],["C:/Users/ACER/Pictures/mydestiny/273536337_788402492117531_8798195010554693138_n.jpg"]]
title = "Object-detection"
description = "An EfficientNetB2 feature extractor computer vision model to classify images of object."
article = "Created by Ryan"
# Create the Gradio demo
demo = gr.Interface(fn=turn_img_into_fileJSON, # mapping function from input to output
                    inputs="pil", # what are the inputs?
                    outputs=[gr.JSON(label="JSON Output"),
                             #gr.Label(num_top_classes=80, label="Predictions"),
                             gr.Number(label="Prediction time (s)")],
                             #gr.outputs.Label(num_top_classes= 80),
                    #examples=path, 
                    title=title,
                    description=description,
                    article=article,
                    live = True)
#demo.launch()
demo.launch(share=True)