Sync internal code changes of tflite/tfhub.

Mostly contributed by Yiqi Li (yuqili@google.com)

efficientdet/README.md

@@ -238,7 +238,7 @@ You can run inference for a video and show the results online:
     // Run eval.
     !python main.py --mode=eval  \
         --model_name=${MODEL}  --model_dir=${CKPT_PATH}  \
-        --validation_file_pattern=tfrecord/val*  \
+        --val_file_pattern=tfrecord/val*  \
         --val_json_file=annotations/instances_val2017.json
 
 You can also run eval on test-dev set with the following command:
@@ -259,7 +259,7 @@ You can also run eval on test-dev set with the following command:
     # Also, test-dev has 20288 images rather than val 5000 images.
     !python main.py --mode=eval  \
         --model_name=${MODEL}  --model_dir=${CKPT_PATH}  \
-        --validation_file_pattern=tfrecord/testdev*  \
+        --val_file_pattern=tfrecord/testdev*  \
         --testdev_dir='testdev_output' --eval_samples=20288
     # Now you can submit testdev_output/detections_test-dev2017_test_results.json to
     # coco server: https://competitions.codalab.org/competitions/20794#participate
@@ -288,8 +288,8 @@ Create a config file for the PASCAL VOC dataset called voc_config.yaml and put t
 Finetune needs to use --ckpt rather than --backbone_ckpt.
 
     !python main.py --mode=train_and_eval \
-        --training_file_pattern=tfrecord/pascal*.tfrecord \
-        --validation_file_pattern=tfrecord/pascal*.tfrecord \
+        --train_file_pattern=tfrecord/pascal*.tfrecord \
+        --val_file_pattern=tfrecord/pascal*.tfrecord \
         --model_name=efficientdet-d0 \
         --model_dir=/tmp/efficientdet-d0-finetune  \
         --ckpt=efficientdet-d0  \
@@ -326,8 +326,8 @@ Download efficientdet coco checkpoint.
 Finetune needs to use --ckpt rather than --backbone_ckpt.
 
     python main.py --mode=train \
-        --training_file_pattern=tfrecord/pascal*.tfrecord \
-        --validation_file_pattern=tfrecord/pascal*.tfrecord \
+        --train_file_pattern=tfrecord/pascal*.tfrecord \
+        --val_file_pattern=tfrecord/pascal*.tfrecord \
         --model_name=efficientdet-d0 \
         --model_dir=/tmp/efficientdet-d0-finetune  \
         --ckpt=efficientdet-d0  \
@@ -358,7 +358,7 @@ To train this model on Cloud TPU, you will need:
 Then train the model:
 
     !export PYTHONPATH="$PYTHONPATH:/path/to/models"
-    !python main.py --tpu=TPU_NAME --training_file_pattern=DATA_DIR/*.tfrecord --model_dir=MODEL_DIR --strategy=tpu
+    !python main.py --tpu=TPU_NAME --train_file_pattern=DATA_DIR/*.tfrecord --model_dir=MODEL_DIR --strategy=tpu
 
     # TPU_NAME is the name of the TPU node, the same name that appears when you run gcloud compute tpus list, or ctpu ls.
     # MODEL_DIR is a GCS location (a URL starting with gs:// where both the GCE VM and the associated Cloud TPU have write access.

efficientdet/dataloader_test.py

@@ -13,63 +13,19 @@
 # limitations under the License.
 # ==============================================================================
 """Data loader and processing test cases."""
-import os
-import tempfile
+
 import tensorflow as tf
 
 import dataloader
 import hparams_config
-from dataset import tfrecord_util
+from brain_automl.efficientdet import test_util
+
 from keras import anchors
 from object_detection import tf_example_decoder
 
 
 class DataloaderTest(tf.test.TestCase):
 
-  def _make_fake_tfrecord(self):
-    tfrecord_path = os.path.join(tempfile.mkdtemp(), 'test.tfrecords')
-    writer = tf.io.TFRecordWriter(tfrecord_path)
-    encoded_jpg = tf.io.encode_jpeg(tf.ones([512, 512, 3], dtype=tf.uint8))
-    example = tf.train.Example(
-        features=tf.train.Features(
-            feature={
-                'image/height':
-                    tfrecord_util.int64_feature(512),
-                'image/width':
-                    tfrecord_util.int64_feature(512),
-                'image/filename':
-                    tfrecord_util.bytes_feature('test_file_name.jpg'.encode(
-                        'utf8')),
-                'image/source_id':
-                    tfrecord_util.bytes_feature('123456'.encode('utf8')),
-                'image/key/sha256':
-                    tfrecord_util.bytes_feature('qwdqwfw12345'.encode('utf8')),
-                'image/encoded':
-                    tfrecord_util.bytes_feature(encoded_jpg.numpy()),
-                'image/format':
-                    tfrecord_util.bytes_feature('jpeg'.encode('utf8')),
-                'image/object/bbox/xmin':
-                    tfrecord_util.float_list_feature([0.1]),
-                'image/object/bbox/xmax':
-                    tfrecord_util.float_list_feature([0.1]),
-                'image/object/bbox/ymin':
-                    tfrecord_util.float_list_feature([0.2]),
-                'image/object/bbox/ymax':
-                    tfrecord_util.float_list_feature([0.2]),
-                'image/object/class/text':
-                    tfrecord_util.bytes_list_feature(['test'.encode('utf8')]),
-                'image/object/class/label':
-                    tfrecord_util.int64_list_feature([1]),
-                'image/object/difficult':
-                    tfrecord_util.int64_list_feature([]),
-                'image/object/truncated':
-                    tfrecord_util.int64_list_feature([]),
-                'image/object/view':
-                    tfrecord_util.bytes_list_feature([]),
-            }))
-    writer.write(example.SerializeToString())
-    return tfrecord_path
-
   def test_parser(self):
     tf.random.set_seed(111111)
     params = hparams_config.get_detection_config('efficientdet-d0').as_dict()
@@ -81,7 +37,7 @@ def test_parser(self):
     anchor_labeler = anchors.AnchorLabeler(input_anchors, params['num_classes'])
     example_decoder = tf_example_decoder.TfExampleDecoder(
         regenerate_source_id=params['regenerate_source_id'])
-    tfrecord_path = self._make_fake_tfrecord()
+    tfrecord_path = test_util.make_fake_tfrecord(self.get_temp_dir())
     dataset = tf.data.TFRecordDataset([tfrecord_path])
     value = next(iter(dataset))
     reader = dataloader.InputReader(tfrecord_path, True)

efficientdet/dataset/create_pascal_tfrecord.py

@@ -33,19 +33,6 @@
 
 from dataset import tfrecord_util
 
-flags.DEFINE_string('data_dir', '', 'Root directory to raw PASCAL VOC dataset.')
-flags.DEFINE_string('set', 'train', 'Convert training set, validation set or '
-                    'merged set.')
-flags.DEFINE_string('annotations_dir', 'Annotations',
-                    '(Relative) path to annotations directory.')
-flags.DEFINE_string('year', 'VOC2007', 'Desired challenge year.')
-flags.DEFINE_string('output_path', '', 'Path to output TFRecord and json.')
-flags.DEFINE_string('label_map_json_path', None,
-                    'Path to label map json file with a dictionary.')
-flags.DEFINE_boolean('ignore_difficult_instances', False, 'Whether to ignore '
-                     'difficult instances')
-flags.DEFINE_integer('num_shards', 100, 'Number of shards for output file.')
-flags.DEFINE_integer('num_images', None, 'Max number of imags to process.')
 FLAGS = flags.FLAGS
 
 SETS = ['train', 'val', 'trainval', 'test']
@@ -79,6 +66,24 @@
 GLOBAL_ANN_ID = 0  # global annotation id.
 
 
+def define_flags():
+  """Define the flags."""
+  flags.DEFINE_string('data_dir', '',
+                      'Root directory to raw PASCAL VOC dataset.')
+  flags.DEFINE_string('set', 'train', 'Convert training set, validation set or '
+                      'merged set.')
+  flags.DEFINE_string('annotations_dir', 'Annotations',
+                      '(Relative) path to annotations directory.')
+  flags.DEFINE_string('year', 'VOC2007', 'Desired challenge year.')
+  flags.DEFINE_string('output_path', '', 'Path to output TFRecord and json.')
+  flags.DEFINE_string('label_map_json_path', None,
+                      'Path to label map json file with a dictionary.')
+  flags.DEFINE_boolean('ignore_difficult_instances', False, 'Whether to ignore '
+                       'difficult instances')
+  flags.DEFINE_integer('num_shards', 100, 'Number of shards for output file.')
+  flags.DEFINE_integer('num_images', None, 'Max number of imags to process.')
+
+
 def get_image_id(filename):
   """Convert a string to a integer."""
   # Warning: this function is highly specific to pascal filename!!
@@ -101,10 +106,9 @@ def get_ann_id():
 
 
 def dict_to_tf_example(data,
-                       dataset_directory,
+                       images_dir,
                        label_map_dict,
                        ignore_difficult_instances=False,
-                       image_subdirectory='JPEGImages',
                        ann_json_dict=None):
   """Convert XML derived dict to tf.Example proto.
 
@@ -114,12 +118,10 @@ def dict_to_tf_example(data,
   Args:
     data: dict holding PASCAL XML fields for a single image (obtained by running
       tfrecord_util.recursive_parse_xml_to_dict)
-    dataset_directory: Path to root directory holding PASCAL dataset
+    images_dir: Path to the directory holding raw images.
     label_map_dict: A map from string label names to integers ids.
     ignore_difficult_instances: Whether to skip difficult instances in the
       dataset  (default: False).
-    image_subdirectory: String specifying subdirectory within the PASCAL dataset
-      directory holding the actual image data.
     ann_json_dict: annotation json dictionary.
 
   Returns:
@@ -128,8 +130,7 @@ def dict_to_tf_example(data,
   Raises:
     ValueError: if the image pointed to by data['filename'] is not a valid JPEG
   """
-  img_path = os.path.join(data['folder'], image_subdirectory, data['filename'])
-  full_path = os.path.join(dataset_directory, img_path)
+  full_path = os.path.join(images_dir, data['filename'])
   with tf.io.gfile.GFile(full_path, 'rb') as fid:
     encoded_jpg = fid.read()
   encoded_jpg_io = io.BytesIO(encoded_jpg)
@@ -297,9 +298,10 @@ def main(_):
       xml = etree.fromstring(xml_str)
       data = tfrecord_util.recursive_parse_xml_to_dict(xml)['annotation']
 
+      img_dir = os.path.join(FLAGS.data_dir, data['folder'], 'JPEGImages')
       tf_example = dict_to_tf_example(
           data,
-          FLAGS.data_dir,
+          img_dir,
           label_map_dict,
           FLAGS.ignore_difficult_instances,
           ann_json_dict=ann_json_dict)
@@ -316,4 +318,5 @@ def main(_):
 
 
 if __name__ == '__main__':
+  define_flags()
   app.run(main)

efficientdet/dataset/create_pascal_tfrecord_test.py

@@ -73,8 +73,9 @@ def test_dict_to_tf_example(self):
         'notperson': 2,
     }
 
-    example = create_pascal_tfrecord.dict_to_tf_example(
-        data, self.get_temp_dir(), label_map_dict, image_subdirectory='')
+    example = create_pascal_tfrecord.dict_to_tf_example(data,
+                                                        self.get_temp_dir(),
+                                                        label_map_dict)
     self._assertProtoEqual(
         example.features.feature['image/height'].int64_list.value, [256])
     self._assertProtoEqual(

efficientdet/dataset/tfrecord_util.py

@@ -72,7 +72,7 @@ def recursive_parse_xml_to_dict(xml):
     Python dictionary holding XML contents.
   """
   if not xml:
-    return {xml.tag: xml.text}
+    return {xml.tag: xml.text if xml.text else ''}
   result = {}
   for child in xml:
     child_result = recursive_parse_xml_to_dict(child)

efficientdet/hparams_config.py

@@ -204,7 +204,7 @@ def default_detection_configs():
   h.max_level = 7
   h.num_scales = 3
   # ratio w/h: 2.0 means w=1.4, h=0.7. Can be computed with k-mean per dataset.
-  h.aspect_ratios = [1.0, 2.0, 0.5]  #[[0.7, 1.4], [1.0, 1.0], [1.4, 0.7]]
+  h.aspect_ratios = [1.0, 2.0, 0.5]  # [[0.7, 1.4], [1.0, 1.0], [1.4, 0.7]]
   h.anchor_scale = 4.0
   # is batchnorm training mode
   h.is_training_bn = True

efficientdet/keras/README.md

@@ -245,7 +245,7 @@ Create a config file for the PASCAL VOC dataset called voc_config.yaml and put t
 Finetune needs to use --pretrained_ckpt.
 
     !python train.py
-        --training_file_pattern=tfrecord/pascal*.tfrecord \
+        --train_file_pattern=tfrecord/pascal*.tfrecord \
         --val_file_pattern=tfrecord/pascal*.tfrecord \
         --val_file_pattern=tfrecord/*.json \
         --model_name=efficientdet-d0 \
@@ -273,7 +273,7 @@ To train this model on Cloud TPU, you will need:
 Then train the model:
 
     !export PYTHONPATH="$PYTHONPATH:/path/to/models"
-    !python train.py --tpu=TPU_NAME --training_file_pattern=DATA_DIR/*.tfrecord --model_dir=MODEL_DIR --strategy=tpu
+    !python train.py --tpu=TPU_NAME --train_file_pattern=DATA_DIR/*.tfrecord --model_dir=MODEL_DIR --strategy=tpu
 
     # TPU_NAME is the name of the TPU node, the same name that appears when you run gcloud compute tpus list, or ctpu ls.
     # MODEL_DIR is a GCS location (a URL starting with gs:// where both the GCE VM and the associated Cloud TPU have write access.