import torch
import torchvision
import torchvision.transforms as transforms
import torch.nn as nn


import os
import sys
import numpy as np
import time
import yaml
import cv2 
import matplotlib.pyplot as plt


"""
Author: Jenö Faist, Paul Judis 
"""

"""
With this File you can use a Video of a 3D Print to Tets your Model.
By Pressing "f" you mark the current frame as a Fail Print 
else the print is considered good
"""

"""
Depending on the Video you must crop it so there a no black lines in it.
You can find out the cropping by just using try and error.
(TODO) IMPLEMENT ADAPTIVE CROPING 
"""

def crop_frame(image):
    cromped_image = image[200:1080-200,620:1920-740,:] #image[:,250:1920-250,:] #image[:,470:1920-700,:] image[:,250:1920-250,:]#image[170:1080-500,650:1920-650,:]
    return cromped_image


if __name__ == '__main__':

    absolutepath = os.path.dirname(__file__)
    video_PATH = absolutepath+'/TEST_VIDEOS/test_4.mp4'
    model_save_PATH = absolutepath+'/COMPLETE_MODELS/3D_DEC_MODEL_MIXR18_18E_64B.pt'


    print("STARTING CNN VIDEO TEST")
    print("--------------------")
    print("Cuda Version:       " + torch.version.cuda)
    print("Cuda:               "+str(torch.cuda.is_available()))
    print("GPU:                "+str(torch.cuda.get_device_name()))
    print("Video PATH:         "+video_PATH)
    print("Model PATH:         "+model_save_PATH)
    print("--------------------")
    print("Loading Model...")


    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    """
    Specifing which Model we using for testing the Data this!
    THIS MUST BE CHANGED DEPENDING OF WHICH MODEL YOU ARE USING !!! 
    """
    model  = torchvision.models.resnet18(weights='DEFAULT')
    num_ftrs = model.fc.in_features
    model.fc = nn.Linear(num_ftrs, 2)

    try:
        model.load_state_dict(torch.load(model_save_PATH))
        model.eval()
        print("[✓] Model Loaded [✓]")
    except:
        print("[!?] No Model Found [?!]")
        exit(1)

    model.eval()
    model = model.to(device)

    print("Loading VIDEO")

    cap = None
    try:
        cap = cv2.VideoCapture(video_PATH)
        print("[✓] Video Loaded [✓]")
    except:
        print("[!?] Video couldn't be found [?!]")
        exit(1)

    print("--------------------")
    print("STARTING TEST THE VIDEO 3D PRINT")
    print("--------------------")

    transform = transforms.Compose([   
        transforms.ToTensor(),
        transforms.Resize((256,256)),
    ])

    running_accrucay = 0
    count = 0
    VIDEO_ON = True
    while(VIDEO_ON):
        count += 1
        ret, frame = cap.read()

        frame = crop_frame(frame)

        input = transform(frame)
        img = (input.squeeze()).numpy()
        img = np.transpose(img, (1, 2, 0))

         
        input = input.to(device)
        output = model(input.unsqueeze(0))
        _, preds = torch.max(output, 1)


        cv2.imshow('TEST MODEL PRESS "F" TO MARK AS FAIL',img)
        k =cv2.waitKey(20)
        FAILED_PRINT = False

        if(k==ord("f")):
            FAILED_PRINT = True

        if(preds[0].item() == 1 and FAILED_PRINT):
            running_accrucay +=1
        elif(preds[0].item() == 0 and not FAILED_PRINT):
            running_accrucay +=1

        """
        JUST FOR VISUALS 
        """

        sym = "NO"
        sym_2 = "NO"
        if(preds[0].item() == 1):
            sym = "YES"
        if FAILED_PRINT:
            sym_2 = "YES"

        
        sys.stdout.write("\033[K")
        print("CNN FAIL DETECTED: " + sym + " USER FAIL DETECTED: " + sym_2 + " CURRENT ACCURACY: "+str(int(100*(running_accrucay/count)))+"%", end='\r')
        
        if k == ord('q'):
            VIDEO_ON = False
        

    sys.stdout.write("\033[K")
    print("TESTING VIDEO FINISHED ACCURACY: "+str(int(100*(running_accrucay/count)))+"%", end='\r')
    cap.release()
    cv2.destroyAllWindows()