checkdata

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sun Apr  7 10:10:12 2019

@author: fanzy
"""


import matplotlib.pyplot as plt
import matplotlib.patches as patches
import numpy as np
from utils import shapeData as dataSet
from config import Config


config = Config()
dataset = dataSet([64,64], config=config)

def data_Gen(dataset,num_batch,batch_size,config):
    for _ in range(num_batch):
        images=[]
        bboxes=[]
        class_ids=[]
        target_matches=[]
        rpn_bboxes=[]
        for i in range(batch_size):
            image, bbox, class_id, rpn_match, rpn_bbox, anchors = data = dataset.load_data()
            pad_num = config.max_gt_obj - bbox.shape[0]
            pad_box = np.zeros((pad_num, 4))
            pad_ids = np.zeros((pad_num, 1))
            bbox = np.concatenate([bbox, pad_box], axis=0)
            class_id = np.concatenate([class_id, pad_ids], axis=0)
        
            images.append(image)
            bboxes.append(bbox)
            class_ids.append(class_id)
            target_matches.append(rpn_match)
            rpn_bboxes.append(rpn_bbox)
        images = np.concatenate(images, 0).reshape(batch_size, config.image_size[0],config.image_size[1] , 3)
        bboxes = np.concatenate(bboxes, 0).reshape(batch_size, -1 , 4)
        class_ids = np.concatenate(class_ids, 0).reshape(batch_size, -1 )
        target_matches = np.concatenate(target_matches, 0).reshape(batch_size, -1 , 1)
        rpn_bboxes = np.concatenate(rpn_bboxes, 0).reshape(batch_size, -1 , 4)
        yield [images, bboxes, class_ids, target_matches, rpn_bboxes,anchors],[]
    
dataGen = data_Gen(dataset, 35000, 1, config)

#拿出一个数据，数据里面包括
#images－－图片　bbox－－即ground truth
#class－－ground truth的类别(哪种物体0,1,2,3,4....)
#input_rpn_matchs－－anchor的类别label(-1,0,1)
#                  　IOU>0.7--label=1(正类), IOU<0.1--label=-1(负类), 0.1<IOU<0.7--label=0(学不到有用信息不参与训练)
#                  　只有１和－１label的anchor参与训练RPN分类，label=1的框是有限的，而＝-1的框很多，所以限制－１和１的框的数量和为Ｎ=100
#input_rpn_bbox－－label=1的anchor,只有label=1的参与回归训练(bbox regression),anchors－576个anchor
data_test=next(dataGen)
images=data_test[0][0]
bboxes=data_test[0][1]
class_ids=data_test[0][2]
input_rpn_matchs=data_test[0][3]
input_rpn_bbox=data_test[0][4]
anchors=data_test[0][5]

plt.imshow(images[0])
axs=plt.gca()

#取出label=1的anchor并画出来,数目为IOU>0.7的anchor数(比如６)
idxposi=np.where(input_rpn_matchs==1)
idx_posi=idxposi[1]
for i in range(idx_posi.shape[0]):
    box=anchors[idx_posi[i]]
    rec=patches.Rectangle((box[0],box[1]),box[2]-box[0],box[3]-box[1],facecolor='none',edgecolor='r')
    axs.add_patch(rec)

#取出label=-1的anchor并画出来，数目=N-IOU>0.7的anchor数(Ｎ－６)
idxnega=np.where(input_rpn_matchs==-1)
idex_nega=idxnega[1]  
for i in range(idex_nega.shape[0]):
    box=anchors[idex_nega[i]]
    rec=patches.Rectangle((box[0],box[1]),box[2]-box[0],box[3]-box[1],facecolor='none',edgecolor='b')
    axs.add_patch(rec)