def _test_preprocess_true_boxes():
dataset = Dataset('train')
# anno = '/home/ylqi007/work/DATA/VOC2007/test/JPEGImages/004538.jpg 215,35,308,224,14 41,77,403,367,12'
anno = '/home/ylqi007/work/DATA/VOC2007/test/JPEGImages/004538.jpg 215,35,308,224,14'
image, bboxes = dataset.parse_annotation(annotation=anno)
print("bboxes after dataset.parse_annotation()\n", bboxes)
dataset.preprocess_true_boxes(bboxes)
def predict(self, image, annotation):
dataset_obj = Dataset('test')
dataset_obj.train_input_size = cfg.TEST.INPUT_SIZE
dataset_obj.strides = np.array(cfg.YOLO.STRIDES)
dataset_obj.train_output_sizes = dataset_obj.train_input_size // dataset_obj.strides
org_image = np.copy(image)
org_h, org_w, _ = org_image.shape
image_data = utils.image_preporcess(image,
[self.input_size, self.input_size])
image_data = image_data[np.newaxis, ...]
_, gt_bboxes = dataset_obj.parse_annotation(annotation)
label_sbbox, label_mbbox, label_lbbox, sbboxes, mbboxes, lbboxes = dataset_obj.preprocess_true_boxes(
gt_bboxes)
label_sbbox = label_sbbox[np.newaxis, ...]
label_mbbox = label_mbbox[np.newaxis, ...]
label_lbbox = label_lbbox[np.newaxis, ...]
sbboxes = sbboxes[np.newaxis, ...]
mbboxes = mbboxes[np.newaxis, ...]
lbboxes = lbboxes[np.newaxis, ...]
pred_sbbox, pred_mbbox, pred_lbbox, loss = self.sess.run(
[self.pred_sbbox, self.pred_mbbox, self.pred_lbbox, self.loss],
feed_dict={
self.input_data: image_data,
self.label_sbbox: label_sbbox,
self.label_mbbox: label_mbbox,
self.label_lbbox: label_lbbox,
self.true_sbboxes: sbboxes,
self.true_mbboxes: mbboxes,
self.true_lbboxes: lbboxes,
self.trainable: False
})
pred_bbox = np.concatenate([
np.reshape(pred_sbbox, (-1, 5 + self.num_classes)),
np.reshape(pred_mbbox, (-1, 5 + self.num_classes)),
np.reshape(pred_lbbox, (-1, 5 + self.num_classes))
],
axis=0)
bboxes = utils.postprocess_boxes(pred_bbox, (org_h, org_w),
self.input_size, self.score_threshold)
bboxes = utils.nms(bboxes, self.iou_threshold)
return bboxes, loss
