Use TensorFlow Object Detection API to achieve transfer learning

Log

---

Test environment:

GPU:  TITAN Xp
CPU:  Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
OS: Ubuntu 16.04
Anaconda: conda 4.6.11
Python: 3.5.6
Tensorflow: 1.10.0

data preparation

Update:

About generate TF record, the Internet to find another METHODS: Run the following script to the training data set and test data set, and get train.record val.record

# xml2csv.py

import os
import glob
import pandas as pd
import xml.etree.ElementTree as ET

os.chdir('/root/proj_emotor/data/VOCdevkit/VOC2012/Annotations')
path ='/root/proj_emotor/data/VOCdevkit/VOC2012/Annotations'

def xml_to_csv(path):
    xml_list = []
    for xml_file in glob.glob(path + '/*.xml'):
        tree = ET.parse(xml_file)
        root = tree.getroot()
        for member in root.findall('object'):
            value = (root.find('filename').text,
                     int(root.find('size')[0].text),
                     int(root.find('size')[1].text),
                     member[0].text,
                     int(member[4][0].text),
                     int(member[4][1].text),
                     int(member[4][2].text),
                     int(member[4][3].text)
                     )
            xml_list.append(value)
    column_name = ['filename', 'width', 'height', 'class', 'xmin', 'ymin', 'xmax', 'ymax']
    xml_df = pd.DataFrame(xml_list, columns=column_name)
    return xml_df


def main():
    image_path = path
    xml_df = xml_to_csv(image_path)
    xml_df.to_csv('emotor_train.csv', index=None)
    print('Successfully converted xml to csv.')


main()

---

# generate_tfrecord.py
# -*- coding: utf-8 -*-
"""
Usage:
  # From tensorflow/models/
  # Create train data:
  python generate_tfrecord.py --csv_input=data/tv_vehicle_labels.csv  --output_path=train.record
  # Create test data:
  python generate_tfrecord.py --csv_input=data/test_labels.csv  --output_path=test.record
"""
import os
import io
import pandas as pd
import tensorflow as tf

from PIL import Image
from object_detection.utils import dataset_util
from collections import namedtuple, OrderedDict

os.chdir('/root/proj_emotor/data')

flags = tf.app.flags
flags.DEFINE_string('csv_input', '', 'Path to the CSV input')
flags.DEFINE_string('output_path', '', 'Path to output TFRecord')
FLAGS = flags.FLAGS


# TO-DO replace this with label map
def class_text_to_int(row_label):
    if row_label == 'emotor':     # 需改动
        return 1
    else:
        None


def split(df, group):
    data = namedtuple('data', ['filename', 'object'])
    gb = df.groupby(group)
    return [data(filename, gb.get_group(x)) for filename, x in zip(gb.groups.keys(), gb.groups)]


def create_tf_example(group, path):
    with tf.gfile.GFile(os.path.join(path, '{}'.format(group.filename)), 'rb') as fid:
        encoded_jpg = fid.read()
    encoded_jpg_io = io.BytesIO(encoded_jpg)
    image = Image.open(encoded_jpg_io)
    width, height = image.size

    filename = group.filename.encode('utf8')
    image_format = b'jpg'
    xmins = []
    xmaxs = []
    ymins = []
    ymaxs = []
    classes_text = []
    classes = []

    for index, row in group.object.iterrows():
        xmins.append(row['xmin'] / width)
        xmaxs.append(row['xmax'] / width)
        ymins.append(row['ymin'] / height)
        ymaxs.append(row['ymax'] / height)
        classes_text.append(row['class'].encode('utf8'))
        classes.append(class_text_to_int(row['class']))

    tf_example = tf.train.Example(features=tf.train.Features(feature={
        'image/height': dataset_util.int64_feature(height),
        'image/width': dataset_util.int64_feature(width),
        'image/filename': dataset_util.bytes_feature(filename),
        'image/source_id': dataset_util.bytes_feature(filename),
        'image/encoded': dataset_util.bytes_feature(encoded_jpg),
        'image/format': dataset_util.bytes_feature(image_format),
        'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
        'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
        'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
        'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
        'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
        'image/object/class/label': dataset_util.int64_list_feature(classes),
    }))
    return tf_example


def main(_):
    writer = tf.python_io.TFRecordWriter(FLAGS.output_path)
    path = os.path.join(os.getcwd(), '/root/proj_emotor/data/VOCdevkit/VOC2012/JPEGImages')         #  需改动
    examples = pd.read_csv(FLAGS.csv_input)
    grouped = split(examples, 'filename')
    for group in grouped:
        tf_example = create_tf_example(group, path)
        writer.write(tf_example.SerializeToString())

    writer.close()
    output_path = os.path.join(os.getcwd(), FLAGS.output_path)
    print('Successfully created the TFRecords: {}'.format(output_path))


if __name__ == '__main__':
    tf.app.run()

---

original

A total of 1700 images are not marked. 800 randomly selected to do preliminary tests.

LabelImg use manual tagging, object label name to emotor.

Picture 1-400 label when the folder name for the result
Picture 401-800 of the folder name for ImageSets
lead to get .xmlthere attributes:

<folder>result</folder>
        <filename>1.jpg</filename>
        <path>F:\result\1.jpg</path>

<folder>ImageSets</folder>
        <filename>405.jpg</filename>
        <path>/home/hzq0/VOC2012/ImageSets/405.jpg</path>

one of them

<folder>ImageSets</folder>

Have an impact at the time of generating TFRecord, as explained later.

All resulting .xmlin Anotationsfolder 1-400 in the picture resultfolder, Picture 401-800 on the ImageSetsfolder.

Execute scripts

import os
import random

trainval_percent = 1
train_percent = 0.5
xmlfilepath = 'VOC2012\VOC2012\Annotations'
txtsavepath = 'VOC2012\VOC2012\ImageSets\Main'
total_xml = os.listdir(xmlfilepath)

num=len(total_xml)
list=range(num)
tv=int(num*trainval_percent)
tr=int(tv*train_percent)
trainval= random.sample(list,tv)
train = random.sample(trainval, tr)

ftrainval = open('VOC2012\VOC2012\ImageSets\Main\\trainval.txt', 'w')
ftest = open('VOC2012\VOC2012\ImageSets\Main\\test.txt', 'w')
ftrain = open('VOC2012\VOC2012\ImageSets\Main\\train.txt', 'w')
fval = open('VOC2012\VOC2012\ImageSets\Main\\val.txt', 'w')

for i in list:
    name = total_xml[i][:-4] + ' '+ '1' + '\n'
    if i in trainval:
        ftrainval.write(name)
        if i in train:
            ftrain.write(name)
        else:
            fval.write(name)
    else:
        ftest.write(name)
        
ftrainval.close()
ftrain.close()
fval.close()
ftest.close()

To create ImageSets/Mainfour .txt, elsewhere execute the script needs to modify the file path in which their own.

When you create tfrecord, we found four .txtused only two. How to modify the back say.

When you are ready you require several folders, copy the script create_pascal_tf_record.py , copy files pascal_label_map.pbtxt to your working directory. And modify pascal_label_map.pbtxtthe file nameemotor_label_map.pbtxt

Folder structure

I called the working directory proj_emotor, engineering construction data preparation part as follows:

+ proj_emotor:
    + data:
        - create_pascal_tf_record.py
        - emotor_label_map.pbtxt
        + VOCdevkit:
            + ImageSets:
                + JPEGImages:
                    - 401~800.jpg
            + result:
                + JPEGImages:
                    - 1~400.jpg
            + VOC2012:
                + Anotations:
                    - .xml
                + ImageSets:
                    + Main:
                        - emotor_train.txt
                        - emotor_val.txt
                + JPEGImages：
                    - all pics

Earlier we do not have to turn over the contents of the configuration files and scripts, the following need to make some changes to them.

First, delete emotor_label_map.pbtxtthe content, amend as follows:

item {
  id: 1
  name: 'emotor'
}

Then, modify create_pascal_tf_record.pythe line 165, the aeroplane_changeemotor_

A close look at the tree, the original of ImageSets/Mainwhich four .txtonly retained train.txtand val.txtand at the front have increased emotor_due to such a change in create_pascal_tf_record.py line 165.

After editing, you can execute the script.

Prior to re-explain why VOCdevkitmore resultand ImageSets.

Because the use of LabelImagetagging when former 400 is in the resultfolder to play, after 400 is ImageSetsin play (this is a mistake, the intention is trying folder name VOC2012of), said earlier get .xmlthere so many words

<folder>ImageSets</folder>

Because of this extra information, resulting in the creation of TFRecordthe time will go to the corresponding folder, find JPEGImages. So later when labeling data, you should be unified picture first into the VOC2012folder and then resort to labels.

Creating TFRecord

#From proj_emotor/data

python create_pascal_tf_record.py \
    --label_map_path=pascal_label_map.pbtxt \
    --data_dir=VOCdevkit --year=VOC2012 --set=train\
    --output_path=emotor_train.record

python create_pascal_tf_record.py \
    --label_map_path=pascal_label_map.pbtxt \
    --data_dir=VOCdevkit --year=VOC2012 --set=val\
    --output_path=emotor_val.record

Note position to execute the command.

After successfully get two files in the data directory:

emotor_train.record
emotor_val.record

Training

Ready to work

First, TensorFlow Model Zoo download the official pre-training model.

I use ssd_mobilenet_v2_cocoin proj_emotor/models/modelunpack.

Now proj_emotorunder the folder structure:

+ proj_emotor
    + data:
        ...
    + models:
        - model_main.py  
        - pipeline.config
        + model：
            + train
            + eval
        + ssd_mobilenet_v2_coco_2018_03_29

Which model_main.py copied from its original position.

pipeline.configAfter extracting from the obtained copy ssd_mobilenet_v2_coco_2018_03_29/pipeline.config.

Modify its contents, modify the content as follows:

model {
  ssd {
    num_classes: 1
    image_resizer {
      fixed_shape_resizer {
        height: 300
        width: 300
      }
    }
    feature_extractor {
      type: "ssd_mobilenet_v2"
      depth_multiplier: 1.0
      min_depth: 16
      conv_hyperparams {
        regularizer {
          l2_regularizer {
            weight: 3.99999989895e-05
          }
        }
        initializer {
          truncated_normal_initializer {
            mean: 0.0
            stddev: 0.0299999993294
          }
        }
        activation: RELU_6
        batch_norm {
          decay: 0.999700009823
          center: true
          scale: true
          epsilon: 0.0010000000475
          train: true
        }
      }
      batch_norm_trainable: true
      use_depthwise: true
    }
    box_coder {
      faster_rcnn_box_coder {
        y_scale: 10.0
        x_scale: 10.0
        height_scale: 5.0
        width_scale: 5.0
      }
    }
    matcher {
      argmax_matcher {
        matched_threshold: 0.5
        unmatched_threshold: 0.5
        ignore_thresholds: false
        negatives_lower_than_unmatched: true
        force_match_for_each_row: true
      }
    }
    similarity_calculator {
      iou_similarity {
      }
    }
    box_predictor {
      convolutional_box_predictor {
        conv_hyperparams {
          regularizer {
            l2_regularizer {
              weight: 3.99999989895e-05
            }
          }
          initializer {
            truncated_normal_initializer {
              mean: 0.0
              stddev: 0.0299999993294
            }
          }
          activation: RELU_6
          batch_norm {
            decay: 0.999700009823
            center: true
            scale: true
            epsilon: 0.0010000000475
            train: true
          }
        }
        min_depth: 0
        max_depth: 0
        num_layers_before_predictor: 0
        use_dropout: false
        dropout_keep_probability: 0.800000011921
        kernel_size: 3
        box_code_size: 4
        apply_sigmoid_to_scores: false
      }
    }
    anchor_generator {
      ssd_anchor_generator {
        num_layers: 6
        min_scale: 0.20000000298
        max_scale: 0.949999988079
        aspect_ratios: 1.0
        aspect_ratios: 2.0
        aspect_ratios: 0.5
        aspect_ratios: 3.0
        aspect_ratios: 0.333299994469
      }
    }
    post_processing {
      batch_non_max_suppression {
        score_threshold: 0.300000011921
        iou_threshold: 0.600000023842
        max_detections_per_class: 100
        max_total_detections: 100
      }
      score_converter: SIGMOID
    }
    normalize_loss_by_num_matches: true
    loss {
      localization_loss {
        weighted_smooth_l1 {
        }
      }
      classification_loss {
        weighted_sigmoid {
        }
      }
      hard_example_miner {
        num_hard_examples: 3000
        iou_threshold: 0.990000009537
        loss_type: CLASSIFICATION
        max_negatives_per_positive: 3
        min_negatives_per_image: 3
      }
      classification_weight: 1.0
      localization_weight: 1.0
    }
  }
}
train_config {
  batch_size: 24
  data_augmentation_options {
    random_horizontal_flip {
    }
  }
  data_augmentation_options {
    ssd_random_crop {
    }
  }
  optimizer {
    rms_prop_optimizer {
      learning_rate {
        exponential_decay_learning_rate {
          initial_learning_rate: 0.00400000018999
          decay_steps: 800720
          decay_factor: 0.949999988079
        }
      }
      momentum_optimizer_value: 0.899999976158
      decay: 0.899999976158
      epsilon: 1.0
    }
  }
  fine_tune_checkpoint: "/root/proj_emotor/model/ssd_mobilenet_v2_coco_2018_03_29/model.ckpt"
  num_steps: 200000
  fine_tune_checkpoint_type: "detection"
}
train_input_reader {
  label_map_path: "/root/proj_emotor/data/emotor_label_map.pbtxt"
  tf_record_input_reader {
    input_path: "/root/proj_emotor/data/emotor_train.record"
  }
}
eval_config {
  num_examples: 8000
  max_evals: 10
  use_moving_averages: false
}
eval_input_reader {
  label_map_path: "/root/proj_emotor/data/emotor_label_map.pbtxt"
  shuffle: false
  num_readers: 1
  tf_record_input_reader {
    input_path: "/root/proj_emotor/data/emotor_val.record"
  }
}

Start training

# From the proj_emotor/models directory
python model_main.py \
    --pipeline_config_path=pipline.config \
    --model_dir=model \
    --num_train_steps=80000 \
    --sample_1_of_n_eval_examples=1 \
    --alsologtostderr

You can use TensorBoard real-time view of the progress during the training exercise.

tensorboard --logdor={PATH TO LOG}

Here is the location of the log position during training log output

After the training, model folder to get the following:

model
├── checkpoint
├── eval
│   └── events.out.tfevents.1563524306.hzq
├── events.out.tfevents.1563523648.hzq
├── export
│   └── Servo_0
│       └── 1563551359
│           ├── saved_model.pb
│           └── variables
│               ├── variables.data-00000-of-00001
│               └── variables.index
├── graph.pbtxt
├── model.ckpt-74690.data-00000-of-00001
├── model.ckpt-74690.index
├── model.ckpt-74690.meta
├── model.ckpt-76420.data-00000-of-00001
├── model.ckpt-76420.index
├── model.ckpt-76420.meta
├── model.ckpt-78144.data-00000-of-00001
├── model.ckpt-78144.index
├── model.ckpt-78144.meta
├── model.ckpt-79957.data-00000-of-00001
├── model.ckpt-79957.index
├── model.ckpt-79957.meta
├── model.ckpt-80000.data-00000-of-00001
├── model.ckpt-80000.index
├── model.ckpt-80000.meta
├── pipeline.config
└── train

Get the model for reasoning

 python export_inference_graph.py \
        --input_type=image_tensor \
        --pipeline_config_path=./model/pipeline.config \
        --trained_checkpoint_prefix=/root/proj_emotor/models/model/model.ckpt-80000 \
        --output_directory=./model/train/

Which --pipeline_config_pathis obtained by training pipeline.config, not for trainingpipeline.config

train
├── checkpoint
    ├── frozen_inference_graph.pb
    ├── model.ckpt.data-00000-of-00001
    ├── model.ckpt.index
    ├── model.ckpt.meta
    ├── pipeline.config
    └── saved_model
        ├── saved_model.pb
        └── variables

train got pb folder for files frozen_inference_graph.pb reasoning.

reasoning

The code for the reasoning:

import numpy as np
import os
import six.moves.urllib as urllib
import sys
import tarfile
import tensorflow as tf
import zipfile

from distutils.version import StrictVersion
from collections import defaultdict
from io import StringIO
from matplotlib import pyplot as plt
from PIL import Image

# This is needed since the notebook is stored in the object_detection folder.
sys.path.append("..")
from object_detection.utils import ops as utils_ops

if StrictVersion(tf.__version__) < StrictVersion('1.9.0'):
  raise ImportError('Please upgrade your TensorFlow installation to v1.9.* or later!')


# This is needed to display the images.
#%matplotlib inline

from object_detection.utils import label_map_util

from object_detection.utils import visualization_utils as vis_util

# What model to download.
MODEL_NAME = '/root/proj_emotor/models/model/train/'
#MODEL_FILE = MODEL_NAME + '.tar.gz'
#DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'

# Path to frozen detection graph. This is the actual model that is used for the object detection.
PATH_TO_FROZEN_GRAPH = MODEL_NAME + 'frozen_inference_graph.pb'

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = os.path.join('/root/proj_emotor/data', 'emotor_label_map.pbtxt')

#opener = urllib.request.URLopener()
#opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
#tar_file = tarfile.open(MODEL_FILE)
#for file in tar_file.getmembers():
#  file_name = os.path.basename(file.name)
#  if 'frozen_inference_graph.pb' in file_name:
#    tar_file.extract(file, os.getcwd())

detection_graph = tf.Graph()
with detection_graph.as_default():
  od_graph_def = tf.GraphDef()
  with tf.gfile.GFile(PATH_TO_FROZEN_GRAPH, 'rb') as fid:
    serialized_graph = fid.read()
    od_graph_def.ParseFromString(serialized_graph)
    tf.import_graph_def(od_graph_def, name='')

category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS, use_display_name=True)

def load_image_into_numpy_array(image):
  (im_width, im_height) = image.size
  return np.array(image.getdata()).reshape(
      (im_height, im_width, 3)).astype(np.uint8)

# For the sake of simplicity we will use only 2 images:
# image1.jpg
# image2.jpg
# If you want to test the code with your images, just add path to the images to the TEST_IMAGE_PATHS.
PATH_TO_TEST_IMAGES_DIR = 'TEST_IMGs'
TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, '{}.jpg'.format(i)) for i in range(1, 100) ]

# Size, in inches, of the output images.
IMAGE_SIZE = (12, 8)

def run_inference_for_single_image(image, graph):
  with graph.as_default():
    with tf.Session() as sess:
      # Get handles to input and output tensors
      ops = tf.get_default_graph().get_operations()
      all_tensor_names = {output.name for op in ops for output in op.outputs}
      tensor_dict = {}
      for key in [
          'num_detections', 'detection_boxes', 'detection_scores',
          'detection_classes', 'detection_masks'
      ]:
        tensor_name = key + ':0'
        if tensor_name in all_tensor_names:
          tensor_dict[key] = tf.get_default_graph().get_tensor_by_name(
              tensor_name)
      if 'detection_masks' in tensor_dict:
        # The following processing is only for single image
        detection_boxes = tf.squeeze(tensor_dict['detection_boxes'], [0])
        detection_masks = tf.squeeze(tensor_dict['detection_masks'], [0])
        # Reframe is required to translate mask from box coordinates to image coordinates and fit the image size.
        real_num_detection = tf.cast(tensor_dict['num_detections'][0], tf.int32)
        detection_boxes = tf.slice(detection_boxes, [0, 0], [real_num_detection, -1])
        detection_masks = tf.slice(detection_masks, [0, 0, 0], [real_num_detection, -1, -1])
        detection_masks_reframed = utils_ops.reframe_box_masks_to_image_masks(
            detection_masks, detection_boxes, image.shape[0], image.shape[1])
        detection_masks_reframed = tf.cast(
            tf.greater(detection_masks_reframed, 0.5), tf.uint8)
        # Follow the convention by adding back the batch dimension
        tensor_dict['detection_masks'] = tf.expand_dims(
            detection_masks_reframed, 0)
      image_tensor = tf.get_default_graph().get_tensor_by_name('image_tensor:0')

      # Run inference
      output_dict = sess.run(tensor_dict,
                             feed_dict={image_tensor: np.expand_dims(image, 0)})

      # all outputs are float32 numpy arrays, so convert types as appropriate
      output_dict['num_detections'] = int(output_dict['num_detections'][0])
      output_dict['detection_classes'] = output_dict[
          'detection_classes'][0].astype(np.uint8)
      output_dict['detection_boxes'] = output_dict['detection_boxes'][0]
      output_dict['detection_scores'] = output_dict['detection_scores'][0]
      if 'detection_masks' in output_dict:
        output_dict['detection_masks'] = output_dict['detection_masks'][0]
  return output_dict

for image_path in TEST_IMAGE_PATHS:
  image = Image.open(image_path)
  # the array based representation of the image will be used later in order to prepare the
  # result image with boxes and labels on it.
  image_np = load_image_into_numpy_array(image)
  # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
  image_np_expanded = np.expand_dims(image_np, axis=0)
  # Actual detection.
  output_dict = run_inference_for_single_image(image_np, detection_graph)
  # Visualization of the results of a detection.
  vis_util.visualize_boxes_and_labels_on_image_array(
      image_np,
      output_dict['detection_boxes'],
      output_dict['detection_classes'],
      output_dict['detection_scores'],
      category_index,
      instance_masks=output_dict.get('detection_masks'),
      use_normalized_coordinates=True,
      line_thickness=8)
  plt.figure(figsize=IMAGE_SIZE)
  plt.imshow(image_np)
  plt.show()

problem

1

google.protobuf.text_format.ParseError: 35:7 : Message type "object_detection.protos.SsdFeatureExtractor" has no field named "batch_norm_trainable"

The delete pipline.config

batch_norm_trainable: true

After, it has been resolved. But I do not know what influence, did not come across before.

Explanation

When the first few trials pipeline.configare copied from the Samples / configs / ssd_mobilenet_v2_coco.config , which do not have this configuration option. This option is what specific role is not known.

2

After the start of training, some parameters could not be initialized using checkpoints

Use `tf.data.Dataset.batch(..., drop_remainder=True)`.
W0718 14:56:12.331554 139700418553600 variables_helper.py:141] Variable [FeatureExtractor/MobilenetV2/Conv_1/BatchNorm/beta] is available in checkpoint, but has an incompatible shape with model variable. Checkpoint shape: [[1280]], model variable shape: [[256]]. This variable will not be initialized from the checkpoint.
W0718 14:56:12.331802 139700418553600 variables_helper.py:141] Variable [FeatureExtractor/MobilenetV2/Conv_1/BatchNorm/gamma] is available in checkpoint, but has an incompatible shape with model variable. Checkpoint shape: [[1280]], model variable shape: [[256]]. This variable will not be initialized from the checkpoint.
W0718 14:56:12.331923 139700418553600 variables_helper.py:141] Variable [FeatureExtractor/MobilenetV2/Conv_1/BatchNorm/moving_mean] is available in checkpoint, but has an incompatible shape with model variable. Checkpoint shape: [[1280]], model variable shape: [[256]]. This variable will not be initialized from the checkpoint.
W0718 14:56:12.332051 139700418553600 variables_helper.py:141] Variable [FeatureExtractor/MobilenetV2/Conv_1/BatchNorm/moving_variance] is available in checkpoint, but has an incompatible shape with model variable. Checkpoint shape: [[1280]], model variable shape: [[256]]. This variable will not be initialized from the checkpoint.
W0718 14:56:12.332157 139700418553600 variables_helper.py:141] Variable [FeatureExtractor/MobilenetV2/Conv_1/weights] is available in checkpoint, but has an incompatible shape with model variable. Checkpoint shape: [[1, 1, 320, 1280]], model variable shape: [[1, 1, 320, 256]]. This variable will not be initialized from the checkpoint.
W0718 14:56:12.336651 139700418553600 variables_helper.py:141] Variable [FeatureExtractor/MobilenetV2/layer_19_1_Conv2d_2_1x1_256/weights] is available in checkpoint, but has an incompatible shape with model variable. Checkpoint shape: [[1, 1, 1280, 256]], model variable shape: [[1, 1, 256, 256]]. This variable will not be initialized from the checkpoint.