deeplabv3+ 跑test数据

た入场券 2023-06-06 10:50 13阅读 0赞

官方代码，虽然有is\_training这个关键字，但是没有test的相关数据生成代码，也没有generator代码。如果有的话那就太好了，告诉我。

如果你看过去哦之前的博客，你听的明白，不明白也没关系，下面告诉你怎么改。

# 改点1，build\_voc2012\_data.py #

毕竟test没图片，我就偷个懒，造个图片，放进tfrecord即可。代码极其简单。不多说

if FLAGS.is_training:
                seg_filename = os.path.join(
                    semantic_segmentation_folder,
                    filenames[i] + '.' + FLAGS.label_format)
                seg_data = tf.gfile.FastGFile(seg_filename, 'rb').read()
                seg_height, seg_width = label_reader.read_image_dims(seg_data)
            else :
                seg_filename = os.path.join(
                    semantic_segmentation_folder,
                    filenames[i] + '.' + FLAGS.label_format)
                seg = np.full([height,width],21,dtype=np.int8)
                img = Image.fromarray(seg)
                img.save(seg_filename)
                seg_data = tf.gfile.FastGFile(seg_filename, 'rb').read()
                seg_height, seg_width = label_reader.read_image_dims(seg_data)

# 改点2，data\_generator.py #

_PASCAL_VOC_SEG_INFORMATION = DatasetDescriptor(
        splits_to_sizes={
            'train': 1464,
            'train_aug': 10582,
            'trainval': 2913,
            'val': 1449,
            'test':1456,
        },

函数

def _preprocess_image(self, sample):
        """Preprocesses the image and label.
    
        Args:
          sample: A sample containing image and label.
    
        Returns:
          sample: Sample with preprocessed image and label.
    
        Raises:
          ValueError: Ground truth label not provided during training.
        """
        image = sample[common.IMAGE]
        if self.split_name != common.TEST_SET:
            label = sample[common.LABELS_CLASS]
        else:
            label=None