from __future__ import division

from models import *
from utils.utils import *
from utils.datasets import *
from utils.parse_config import *

import os
import sys
import time
import datetime
import argparse
import tqdm
import cv2
from PIL import Image

import torch
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision import transforms
from torch.autograd import Variable
import torch.nn.functional as F
import torch.optim as optim
import time
import numpy as np 

# 
def preprocess(img):
    h, w, c = img.shape
    dim_diff = np.abs(h - w)
    pad1, pad2 = dim_diff // 2, dim_diff - dim_diff // 2
    pad = ((pad1, pad2), (0,0), (0,0)) if h <= w else ((0,0), (pad1, pad2), (0,0))
    img = np.pad(img, pad, 'constant', constant_values=0)



    print (img.shape)
    dst_img = img
    return dst_img

if __name__ == '__main__':
    # torch 
    device = torch.device("cuda")

    # data setting
    data_config = parse_data_config('config/oxfordhand.data')
    valid_path = data_config["valid"]

    # class name
    class_names = load_classes(data_config["names"])

    # initiate model
    model = Darknet('config/prune_yolov3-hand.cfg').to(device)
    model.load_state_dict(torch.load('weights/prune_yolov3_ckpt.pth'))

    # model test
    model.eval()
    Tensor = torch.cuda.FloatTensor


    cap = cv2.VideoCapture(0)
    w_cap = cv2.VideoWriter('./out.avi', cv2.VideoWriter_fourcc(*'XVID'), 25, (416, 416)) 
    while (cap.isOpened()):
        ret, frame = cap.read()
        if ret == True:
            #frame = cv2.imread('/home/yehao/YOLOv3-model-pruning/test.jpg')
            frame = frame[:, :, ::-1]

            """
            standart method
            """
            #frame = Image.open('/home/yehao/YOLOv3-model-pruning/test.jpg')
            #frame = np.array(frame)


            #print (frame.shape)
            start = time.time()
            # do pad
            #print ('origin img: ', frame[100, 200:250, 0])
            dst_frame = preprocess(frame)	
            print (dst_frame.shape)		
            #print ('pad img: ', dst_frame[100, 200:250, 0])
            show_frame = cv2.resize(dst_frame[:, :, ::-1], (416, 416))

            tensor = transforms.ToTensor()(dst_frame)
            print (tensor.shape)
            #print ('to tensor img: ', tensor[0, 100, 200:250])
            tensor = F.interpolate(tensor.unsqueeze(0), size=416, mode="nearest")
            print (tensor.shape)
            
            
            #imgs = F.interpolate(tensor, size=416, mode="nearest")
            #print (tensor)
            imgs = Variable(tensor.type(Tensor), requires_grad=False)
            print (imgs.shape)

            
            with torch.no_grad():
                outputs = model(imgs)
                outputs = non_max_suppression(outputs, conf_thres=0.01, nms_thres=0.3)

            end = time.time()
            print ('cost time: %.2f s'%(end - start))
            #print (outputs)
            print (type(outputs))
            if str(outputs[0]) == 'None':
                print ('empty')
            else:
                print (outputs[0])
                for i in range(len(outputs[0])):
                    coor = outputs[0][i].tolist()
                    #print (coor)
                    #print (type(coor))
                    x1 = int(coor[0])
                    y1 = int(coor[1])
                    x2 = int(coor[2])
                    y2 = int(coor[3])   
                    conf = coor[4]
                    if conf > 0.6:  
                        print (x1, y1, x2, y2, conf)
                        cv2.rectangle(show_frame, (x1, y1), (x2, y2), (0, 0, 255), 2)
            
            #last = cv2.resize(show_frame, (800, 800))
            w_cap.write(show_frame)
            cv2.imshow('frame', show_frame)
            #cv2.waitKey(10000)
            cv2.waitKey(5)