import os
import time
import glob
import sys
import numpy as np
import openslide
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
from mpl_toolkits.axes_grid1 import make_axes_locatable
import cv2
kernel = np.ones((5,5), np.uint8) 

from PIL import Image
sys.path.append('..')
from DigiPathAI.Segmentation import getSegmentation

digestpath_imgs = ['../examples/colon-cancer-1.tiff']

paip_imgs       = ['../examples/examples/tcga/liver-1.svs']

tcga_imgs       = ['../examples/examples/tcga/TCGA-CM.svs']

camelyon_imgs   = ['../examples/Camelyon_16_Test_samples/camelyon-1.tif']


models = ['dense', 'inception', 'deeplabv3', 'ensemble', 'epistemic']

def iou(gt, mask):
    gt = np.uint8(np.array(gt>0.1))
    mask = np.uint8(np.array(mask>0.1))
    nr = np.sum(gt*mask)*2.0
    dr = np.sum(gt + mask)*1.0
    return nr/dr 

for path in tcga_imgs:
  ext = os.path.splitext(path)[1]
  base_path = os.path.splitext(path)[0]

  print (ext, base_path, base_path[:-5])
  quick = True
  tta_list = ['FLIP_LEFT_RIGHT', 'ROTATE_90'] #, 'ROTATE_180', 'ROTATE_270']
  for model in models:

    print (model, quick, path, "======================================")
    if model == 'ensemble': 
      quick = False
    elif model == 'epistemic':
      quick = False
      tta_list = None
    """
    getSegmentation(path, 
          patch_size  = 512, 
          stride_size = 512,
          batch_size  = 4,
          quick       = quick,
          tta_list    = tta_list,
          crf         = False,
          mask_path   = base_path + '-DigiPathAI_{}_mask'.format(model) + '.tiff',
          uncertainty_path   = base_path + '-DigiPathAI_{}_uncertainty'.format(model)+ '.tiff',
          status      = None,
          mask_level = 4,
          model       = model,
          mode        = 'breast')
    
    """
    slide = openslide.OpenSlide(path)
    level = len(slide.level_dimensions) - 1
    img_dimensions = slide.level_dimensions[-1]
    img = np.array(slide.read_region((0,0), level, img_dimensions).convert('RGB'))

    mask = openslide.OpenSlide(base_path + '-DigiPathAI_{}_mask'.format(model) + '.tiff')
    level = np.where([1 if ((dim[0] == img_dimensions[0])*(dim[1] == img_dimensions[1])) else 0 for dim in mask.level_dimensions])[0]
    mask = np.array(mask.read_region((0,0), level, img_dimensions).convert('L'))
    mask = cv2.dilate(mask, kernel, iterations=2) 

    uncertainty = openslide.OpenSlide(base_path + '-DigiPathAI_{}_uncertainty'.format(model) + '.tiff')
    level = np.where([1 if ((dim[0] == img_dimensions[0])*(dim[1] == img_dimensions[1])) else 0 for dim in uncertainty.level_dimensions])[0]
    uncertainty = np.array(uncertainty.read_region((0,0), level, img_dimensions).convert('L'))
    
    gt = openslide.OpenSlide(glob.glob(base_path + '-gt*')[0])
    level = np.where([1 if ((dim[0] == img_dimensions[0])*(dim[1] == img_dimensions[1])) else 0 for dim in gt.level_dimensions])[0]
    gt = np.array(gt.read_region((0,0), level, img_dimensions).convert('L'))*255
    gt = np.array(Image.fromarray(gt).resize(img_dimensions, Image.NEAREST))
    
    mask = np.array(Image.fromarray(mask).resize(img_dimensions, Image.NEAREST))
    uncertainty = np.array(Image.fromarray(uncertainty).resize(img_dimensions))/255.0
    # gt = np.array(Image.open(base_path + 'gt.jpg').convert('L').resize(img_dimensions))
    


    print ("path: {}, model: {}, IoU: {}".format(path, model, iou(gt, mask)))
    

    fig, ax = plt.subplots(1, 4, figsize=(10, 40))
    im_ = ax[0].imshow(img)
    ax[0].set_xticklabels([])
    ax[0].set_yticklabels([])
    ax[0].set_xticks([])
    ax[0].set_yticks([])
    ax[0].set_aspect('equal')
    ax[0].tick_params(bottom='off', top='off', labelbottom='off', right='off', left='off', labelleft='off' )
    

    im_ = ax[1].imshow(img)
    gt_ = ax[1].imshow(gt, alpha = 0.5, cmap='gray')
    ax[1].set_xticklabels([])
    ax[1].set_yticklabels([])
    ax[1].set_xticks([])
    ax[1].set_yticks([])
    ax[1].set_aspect('equal')
    ax[1].tick_params(bottom='off', top='off', labelbottom='off', right='off', left='off', labelleft='off' )
    
    im_ = ax[2].imshow(img)
    mask_ = ax[2].imshow(mask, alpha = 0.5, cmap='gray')
    ax[2].set_xticklabels([])
    ax[2].set_yticklabels([])
    ax[2].set_xticks([])
    ax[2].set_yticks([])
    ax[2].set_aspect('equal')
    ax[2].tick_params(bottom='off', top='off', labelbottom='off', right='off', left='off', labelleft='off' )
    
    im_ = ax[3].imshow(img)
    uncertain_ = ax[3].imshow(uncertainty, alpha = 0.5, cmap=plt.cm.RdBu_r)
    ax[3].set_xticklabels([])
    ax[3].set_yticklabels([])
    ax[3].set_xticks([])
    ax[3].set_yticks([])
    ax[3].set_aspect('equal')
    ax[3].tick_params(bottom='off', top='off', labelbottom='off', right='off', left='off', labelleft='off' )

    cax = fig.add_axes([ax[3].get_position().x1 + 0.01,
              ax[3].get_position().y0,
              0.01,
              ax[3].get_position().y1-ax[3].get_position().y0])
    fig.colorbar(uncertain_, cax=cax)

    plt.savefig(base_path+'DigiPath_Results_{}.png'.format(model), bbox_inches='tight')