Update the codes due to recent change in DeepLab. (#6287)

research/feelvos/common.py

@@ -137,3 +137,27 @@ class VideoModelOptions(common.ModelOptions):
     self.classification_loss = FLAGS.classification_loss
 
     return self
+
+
+def parse_decoder_output_stride():
+  """Parses decoder output stride.
+
+  FEELVOS assumes decoder_output_stride = 4. Thus, this function is created for
+  this particular purpose.
+
+  Returns:
+    An integer specifying the decoder_output_stride.
+
+  Raises:
+    ValueError: If decoder_output_stride is None or contains more than one
+      element.
+  """
+  if FLAGS.decoder_output_stride:
+    decoder_output_stride = [
+        int(x) for x in FLAGS.decoder_output_stride]
+    if len(decoder_output_stride) != 1:
+      raise ValueError('Expect decoder output stride has only one element.')
+    decoder_output_stride = decoder_output_stride[0]
+  else:
+    raise ValueError('Expect flag decoder output stride not to be None.')
+  return decoder_output_stride

research/feelvos/model.py

@@ -359,9 +359,10 @@ def multi_scale_logits_with_nearest_neighbor_matching(
       if model_options.crop_size else tf.shape(images)[2])
 
   # Compute the height, width for the output logits.
-  logits_output_stride = (
-      model_options.decoder_output_stride or model_options.output_stride)
-
+  if model_options.decoder_output_stride:
+    logits_output_stride = min(model_options.decoder_output_stride)
+  else:
+    logits_output_stride = model_options.output_stride
   logits_height = scale_dimension(
       crop_height,
       max(1.0, max(image_pyramid)) / logits_output_stride)

research/feelvos/train.py

@@ -266,7 +266,7 @@ def _build_deeplab(inputs_queue_or_samples, outputs_to_num_classes,
             preceding_frame_label[n, tf.newaxis],
             samples[common.LABEL][n * FLAGS.train_num_frames_per_video,
                                   tf.newaxis],
-            FLAGS.decoder_output_stride,
+            common.parse_decoder_output_stride(),
             reduce_labels=True)
         init_softmax_n = tf.squeeze(init_softmax_n, axis=0)
         init_softmax.append(init_softmax_n)
@@ -610,7 +610,7 @@ def _get_dataset_and_samples(config, train_crop_size, dataset_name,
       is_training=True,
       model_variant=FLAGS.model_variant,
       batch_capacity_factor=FLAGS.batch_capacity_factor,
-      decoder_output_stride=FLAGS.decoder_output_stride,
+      decoder_output_stride=common.parse_decoder_output_stride(),
       first_frame_finetuning=first_frame_finetuning,
       sample_only_first_frame_for_finetuning=
       FLAGS.sample_only_first_frame_for_finetuning,

research/feelvos/utils/embedding_utils.py

@@ -482,20 +482,17 @@ def get_embeddings(images, model_options, embedding_dimension):
       is_training=False)
 
   if model_options.decoder_output_stride is not None:
+    decoder_output_stride = min(model_options.decoder_output_stride)
     if model_options.crop_size is None:
       height = tf.shape(images)[1]
       width = tf.shape(images)[2]
     else:
       height, width = model_options.crop_size
-    decoder_height = model.scale_dimension(
-        height, 1.0 / model_options.decoder_output_stride)
-    decoder_width = model.scale_dimension(
-        width, 1.0 / model_options.decoder_output_stride)
     features = model.refine_by_decoder(
         features,
         end_points,
-        decoder_height=decoder_height,
-        decoder_width=decoder_width,
+        crop_size=[height, width],
+        decoder_output_stride=[decoder_output_stride],
         decoder_use_separable_conv=model_options.decoder_use_separable_conv,
         model_variant=model_options.model_variant,
         is_training=False)
@@ -596,21 +593,20 @@ def get_logits_with_matching(images,
       is_training=is_training,
       fine_tune_batch_norm=fine_tune_batch_norm)
 
-  if model_options.decoder_output_stride is not None:
+  if model_options.decoder_output_stride:
+    decoder_output_stride = min(model_options.decoder_output_stride)
     if model_options.crop_size is None:
       height = tf.shape(images)[1]
       width = tf.shape(images)[2]
     else:
       height, width = model_options.crop_size
-    decoder_height = model.scale_dimension(
-        height, 1.0 / model_options.decoder_output_stride)
-    decoder_width = model.scale_dimension(
-        width, 1.0 / model_options.decoder_output_stride)
+    decoder_height = model.scale_dimension(height, 1.0 / decoder_output_stride)
+    decoder_width = model.scale_dimension(width, 1.0 / decoder_output_stride)
     features = model.refine_by_decoder(
         features,
         end_points,
-        decoder_height=decoder_height,
-        decoder_width=decoder_width,
+        crop_size=[height, width],
+        decoder_output_stride=[decoder_output_stride],
         decoder_use_separable_conv=model_options.decoder_use_separable_conv,
         model_variant=model_options.model_variant,
         weight_decay=weight_decay,

research/feelvos/vis_video.py

@@ -222,7 +222,7 @@ def create_predictions(samples, reference_labels, first_frame_img,
 
   init_labels = tf.squeeze(reference_labels, axis=-1)
   init_softmax = embedding_utils.create_initial_softmax_from_labels(
-      reference_labels, reference_labels, FLAGS.decoder_output_stride,
+      reference_labels, reference_labels, common.parse_decoder_output_stride(),
       reduce_labels=False)
   if FLAGS.save_embeddings:
     decoder_height = tf.shape(init_softmax)[1]
@@ -298,7 +298,7 @@ def create_predictions_fast(samples, reference_labels, first_frame_img,
       first_frame_img[tf.newaxis], model_options, FLAGS.embedding_dimension)
   init_labels = tf.squeeze(reference_labels, axis=-1)
   init_softmax = embedding_utils.create_initial_softmax_from_labels(
-      reference_labels, reference_labels, FLAGS.decoder_output_stride,
+      reference_labels, reference_labels, common.parse_decoder_output_stride(),
       reduce_labels=False)
   init = (init_labels, init_softmax, first_frame_embeddings)