import numpy as np
import sys
from PIL import Image, ImageDraw, ImageEnhance, ImageColor

class Rpn:
    def __init__(self, tf, resnet_model, image_height, image_width, in_channels, mid_channels, base_anchor_size, anchor_ratios, anchor_scales, subsample_rate):
        """
            in_channels for resnet: 512
            mid_channels: 512
            base_anchor_size: 128
            anchor_ratios: [[1,1],[1,2],[2,1]]
            anchor_scales: [1,2,3,4]
            subsample_rate: (image_size//feature_map_size), ex. (1000//32) --> 31
        """
        self.tf = tf
        self.resnet_model = resnet_model
        self.resnet_model_inputs = self.resnet_model.input
        self.resnet_model_feature_map = resnet_model.output
        self.image_height = image_height
        self.image_width = image_width
        self.in_channels = in_channels
        self.mid_channels = mid_channels
        self.base_anchor_size = base_anchor_size
        self.anchor_ratios = anchor_ratios
        self.anchor_scales = anchor_scales
        self.subsample_rate = subsample_rate


        self.anchor_boxes = self.generate_anchor_boxes(self.base_anchor_size, self.anchor_ratios, self.anchor_scales)
        self.anchor_boxes_over_image = self.generate_anchor_boxes_over_image(self.anchor_boxes, self.image_height, self.image_width, self.subsample_rate)

        self.model = self.create_model()

    def get_model(self):
        return self.model

    def create_model(self):
        rpn_conv = self.tf.keras.layers.SeparableConv2D(filters=512, kernel_size=[3, 3], strides=[1, 1], padding="same", data_format="channels_last", activation="relu")(self.resnet_model.output)
        rpn_class_score = self.tf.keras.layers.SeparableConv2D(filters=24, kernel_size=[1, 1], strides=[1, 1], padding="valid", data_format="channels_last", activation="softmax")(rpn_conv)
        rpn_class_score_shape = rpn_class_score.get_shape().as_list()
        rpn_class_score = self.tf.keras.layers.Reshape(rpn_class_score_shape[1:3] + [rpn_class_score_shape[3]//2, 2])(rpn_class_score)

        rpn_bbox_pred = self.tf.keras.layers.SeparableConv2D(filters=48, kernel_size=[1, 1], strides=[1, 1], padding="valid", data_format="channels_last")(rpn_conv)
        rpn_bbox_pred_shape = rpn_bbox_pred.get_shape().as_list()
        rpn_bbox_pred = self.tf.keras.layers.Reshape(rpn_bbox_pred_shape[1:3] + [rpn_bbox_pred_shape[3]//4, 4])(rpn_bbox_pred)

        rpn_model = self.tf.keras.Model(inputs=self.resnet_model.input, outputs=[rpn_class_score, rpn_bbox_pred])

        self.optimizer = self.tf.keras.optimizers.Nadam(0.001)
        #rpn_model.compile(optimizer=self.optimizer, loss=[self.tf.keras.losses.BinaryCrossentropy(),self.tf.keras.losses.Huber()], metrics=['binary_accuracy', 'MeanSquaredError'])
        return rpn_model

    def get_iou(self, anchor_box_predictions, ground_truth_bounding_boxes, giou=False):
        """ Predicted and ground truth bounding box coordinates """
        anchor_box_predictions_center_x, anchor_box_predictions_center_y, anchor_box_predictions_width, anchor_box_predictions_height = anchor_box_predictions[:,:,:,:,0:1], anchor_box_predictions[:,:,:,:,1:2], anchor_box_predictions[:,:,:,:,2:3], anchor_box_predictions[:,:,:,:,3:4]
        anchor_box_predictions_x1 = anchor_box_predictions_center_x - (anchor_box_predictions_width / 2)
        anchor_box_predictions_x2 = anchor_box_predictions_center_x + (anchor_box_predictions_width / 2)
        anchor_box_predictions_y1 = anchor_box_predictions_center_y - (anchor_box_predictions_height / 2)
        anchor_box_predictions_y2 = anchor_box_predictions_center_y + (anchor_box_predictions_height / 2)

        """ For the predicted box ensure x2>x1 and y2>y1: """
        anchor_box_predictions_x1, anchor_box_predictions_x2 = np.minimum(anchor_box_predictions_x1, anchor_box_predictions_x2), np.maximum(anchor_box_predictions_x1, anchor_box_predictions_x2)
        anchor_box_predictions_y1, anchor_box_predictions_y2 = np.minimum(anchor_box_predictions_y1, anchor_box_predictions_y2), np.maximum(anchor_box_predictions_y1, anchor_box_predictions_y2)

        """ Flatten ground truth boxes sorta """
        concat = np.concatenate(ground_truth_bounding_boxes, axis=0)
        ground_truth_bounding_boxes_center_x, ground_truth_bounding_boxes_center_y, ground_truth_bounding_boxes_width, ground_truth_bounding_boxes_height = (concat[:, 0:1]).reshape((1,1,1,1,-1)), (concat[:, 1:2]).reshape((1,1,1,1,-1)), (concat[:, 2:3]).reshape((1,1,1,1,-1)), (concat[:, 3:4]).reshape((1,1,1,1,-1))

        ground_truth_bounding_boxes_x1 = ground_truth_bounding_boxes_center_x - (ground_truth_bounding_boxes_width / 2)
        ground_truth_bounding_boxes_x2 = ground_truth_bounding_boxes_center_x + (ground_truth_bounding_boxes_width / 2)
        ground_truth_bounding_boxes_y1 = ground_truth_bounding_boxes_center_y - (ground_truth_bounding_boxes_height / 2)
        ground_truth_bounding_boxes_y2 = ground_truth_bounding_boxes_center_y + (ground_truth_bounding_boxes_height / 2)



        """ Get areas of boxes """
        anchor_box_predictions_area = (anchor_box_predictions_x2 - anchor_box_predictions_x1) * (anchor_box_predictions_y2 - anchor_box_predictions_y1)
        ground_truth_bounding_boxes_area = (ground_truth_bounding_boxes_x2 - ground_truth_bounding_boxes_x1) * (ground_truth_bounding_boxes_y2 - ground_truth_bounding_boxes_y1)

        """ Calculate intersection between prediction boxes and ground truth boxes """

        x1_intersection, x2_intersection = np.maximum(anchor_box_predictions_x1, ground_truth_bounding_boxes_x1), np.minimum(anchor_box_predictions_x2, ground_truth_bounding_boxes_x2)
        y1_intersection, y2_intersection = np.maximum(anchor_box_predictions_y1, ground_truth_bounding_boxes_y1), np.minimum(anchor_box_predictions_y2, ground_truth_bounding_boxes_y2)


        """ Calculate intersection area """
        intersection = np.where(np.logical_and(x2_intersection > x1_intersection, y2_intersection > y1_intersection), (x2_intersection - x1_intersection) * (y2_intersection - y1_intersection), 0)

        """ Find the coordinates of smallest enclosing box (union) """
        x1_coord, x2_coord = np.minimum(anchor_box_predictions_x1, ground_truth_bounding_boxes_x1), np.maximum(anchor_box_predictions_x2, ground_truth_bounding_boxes_x2)
        y1_coord, y2_coord = np.minimum(anchor_box_predictions_y1, ground_truth_bounding_boxes_y1), np.maximum(anchor_box_predictions_y2, ground_truth_bounding_boxes_y2)

        """ Get area of union box """
        union_area = (x2_coord - x1_coord) * (y2_coord - y1_coord)

        """ Intersection over union """
        iou = intersection / (anchor_box_predictions_area + ground_truth_bounding_boxes_area - intersection)
        Giou = iou - ((union_area-(anchor_box_predictions_area + ground_truth_bounding_boxes_area - intersection))/union_area)

        indices = [len(bb) for bb in ground_truth_bounding_boxes]
        indices_end = [np.sum(indices[:i + 1]) for i in range(len(indices))]
        indices_start = np.insert(indices_end, 0, 0)[:-1]

        """ get iou, and giou with respect to correct bounding boxes for each batch image, also get those correct bounding boxes """
        iou_true = np.zeros(iou.shape[:-1])
        Giou_true = np.zeros(Giou.shape[:-1])
        bounding_boxes_true = np.zeros(anchor_box_predictions.shape)
        for index, (s, e) in enumerate(zip(indices_start, indices_end)):
            iou_true[index] = np.amax(iou[index,:,:,:, s:e], axis=3)
            Giou_true[index] = np.amax(Giou[index,:,:,:, s:e], axis=3)
            if giou is True:
                bounding_boxes_true[index] = (ground_truth_bounding_boxes[index][np.argmax(Giou[index,:,:,:, s:e], axis=3)])[:,:,:,:4]
            else:
                bounding_boxes_true[index] = (ground_truth_bounding_boxes[index][np.argmax(iou[index,:,:,:, s:e], axis=3)])[:,:,:,:4]



        return (Giou_true,bounding_boxes_true) if giou else (iou_true, bounding_boxes_true)


    def train_model(self, inputs, ground_truth_bounding_boxes):
        batch_size = inputs.shape[0]
        predictions = self.model.predict(inputs)
        object_predictions = predictions[0]
        anchor_box_change_predictions = predictions[1]
        anchor_box_predictions = self.anchor_boxes_over_image + anchor_box_change_predictions
        anchor_boxes_giou, anchor_boxes_nearest_bounding_box = self.get_iou(anchor_box_predictions, ground_truth_bounding_boxes, giou=True)
        object_prediction_labels = np.zeros(object_predictions.shape[:-1])
        object_prediction_labels[anchor_boxes_giou > 0.5] += 1
        print(object_predictions.shape, anchor_box_change_predictions.shape, object_prediction_labels.shape)
        print([x.shape for x in self.model.trainable_weights])
        sys.exit()


        mini_batch_sample_size = 256*batch_size
        object_predictions_flat = object_predictions.flatten()
        object_prediction_labels_flat = object_prediction_labels.flatten()
        anchor_box_predictions_flat = anchor_box_predictions.reshape((-1, 4))
        anchor_boxes_nearest_bounding_box_flat = anchor_boxes_nearest_bounding_box.reshape((-1, 4))


        """Shuffle flattened y_pred and y_true all in the same order"""
        random_indices = np.random.choice(object_predictions_flat.shape[0], object_predictions_flat.shape[0], replace=False)
        object_predictions_flat = object_predictions_flat[random_indices]
        object_prediction_labels_flat = object_prediction_labels_flat[random_indices]
        anchor_box_predictions_flat = anchor_box_predictions_flat[random_indices]
        anchor_boxes_nearest_bounding_box_flat = anchor_boxes_nearest_bounding_box_flat[random_indices]

        """ Sort in ascending order from background to foreground """
        ind_sorted = np.argsort(object_prediction_labels_flat)
        object_predictions_flat = object_predictions_flat[ind_sorted]
        object_prediction_labels_flat = object_prediction_labels_flat[ind_sorted]
        anchor_box_predictions_flat = anchor_box_predictions_flat[ind_sorted]
        anchor_boxes_nearest_bounding_box_flat = anchor_boxes_nearest_bounding_box_flat[ind_sorted]

        """ Get 128 background anchors and 128 foreground anchors and merge them into a single batch """
        split_amount = mini_batch_sample_size // 2
        #Background
        background_object_predictions_flat = object_predictions_flat[:split_amount]
        background_object_prediction_labels_flat = object_prediction_labels_flat[:split_amount]

        #Foreground
        foreground_object_predictions_flat = object_predictions_flat[-split_amount:]
        foreground_object_prediction_labels_flat = object_prediction_labels_flat[-split_amount:]
        foreground_anchor_box_predictions_flat = anchor_box_predictions_flat[-split_amount:]
        foreground_anchor_boxes_nearest_bounding_box_flat = anchor_boxes_nearest_bounding_box_flat[-split_amount:]


        #Merge
        final_object_predictions_flat = np.concatenate((background_object_predictions_flat,foreground_object_predictions_flat))
        final_object_prediction_labels_flat = np.concatenate((background_object_prediction_labels_flat, foreground_object_prediction_labels_flat))
        #Take only foreground anchor and the closest ground truth boxes from the mini batch
        final_anchor_box_predictions_flat = foreground_anchor_box_predictions_flat[np.nonzero(foreground_object_prediction_labels_flat)]
        final_anchor_boxes_nearest_bounding_box_flat = foreground_anchor_boxes_nearest_bounding_box_flat[np.nonzero(foreground_object_prediction_labels_flat)]



        #====================================================WHAT DO I DO FROM HERE?????????????????====================================================

        class_binary_loss = self.tf.keras.losses.BinaryCrossentropy
        huber = self.tf.keras.losses.Huber(reduction=self.tf.keras.losses.Reduction.NONE)
        anchor_box_regression_loss = huber(final_anchor_boxes_nearest_bounding_box_flat, final_anchor_box_predictions_flat)
        print(anchor_box_regression_loss.shape)

        #print((tf.keras.optimizers.get_weights()).shape)
        #print([x.shape for x in self.model.trainable_weights])
        lo = class_binary_loss(final_object_prediction_labels_flat, final_object_predictions_flat)
        #print(lo.shape)
        #print(lo.shape)
        g = self.tf.keras.backend.gradients(self.tf.ones([1], tf.int32), self.model.trainable_weights[:-3])
        #self.optimizer.minimize(lo, self.model.trainable_weights[:-3])
        #self.optimizer.minimize(anchor_box_regression_loss, self.model.trainable_weights[:-6]+self.model.trainable_weights[-3:])
        #print(class_binary_loss, anchor_box_regression_loss)


    def generate_anchor_boxes(self, base_anchor_size, anchor_ratios, anchor_scales):
        """
            Every anchor box is different
            Number of anchor boxes: len(anchor_ratios)*len(anchor_scales)
        """

        anchor_boxes = []
        for scale in anchor_scales:
            for aspect_ratio in anchor_ratios:
                height = base_anchor_size*scale*aspect_ratio[0]
                width = base_anchor_size*scale*aspect_ratio[1]
                anchor_box = [width, height]
                anchor_boxes.append(anchor_box)

        return anchor_boxes

    def generate_anchor_boxes_over_image(self, anchor_boxes, image_height, image_width, subsample_rate):
        """
            Returns an array of all the anchor boxes, each with shape [anchor_box_center_x, anchor_box_center_y, width, height]
        """
        anchor_boxes_over_image = np.zeros(( image_height//subsample_rate, image_width//subsample_rate, len(anchor_boxes), 4))
        for row in range(anchor_boxes_over_image.shape[0]):
            for col in range(anchor_boxes_over_image.shape[1]):
                for anchor_idx in range(anchor_boxes_over_image.shape[2]):
                    anchor_boxes_over_image[row,col,anchor_idx] = [(row+1)*subsample_rate, (col+1)*subsample_rate, anchor_boxes[anchor_idx][0], anchor_boxes[anchor_idx][1]]
        return anchor_boxes_over_image