Explicitly import estimator from tensorflow as a separate import instead of

accessing it via tf.estimator and depend on the tensorflow estimator target.

PiperOrigin-RevId: 439395101

Author: hertschuh

tensorflow_graphics/projects/nasa/lib/models.py

@@ -14,6 +14,7 @@
 """Model Implementations."""
 import tensorflow.compat.v1 as tf
 
+from tensorflow.compat.v1 import estimator as tf_estimator
 from tensorflow_graphics.projects.nasa.lib import model_utils
 
 tf.disable_eager_execution()
@@ -37,7 +38,7 @@ def nasa(hparams):
   sample_bbox = hparams.sample_bbox
 
   def _model_fn(features, labels, mode, params=None):
-    is_training = (mode == tf.estimator.ModeKeys.TRAIN)
+    is_training = (mode == tf_estimator.ModeKeys.TRAIN)
     batch_size = features['point'].shape[0]
     n_sample_frames = features['point'].shape[1]
     accum_size = batch_size * n_sample_frames
@@ -116,7 +117,7 @@ def _model_fn(features, labels, mode, params=None):
       train_op = optimizer.minimize(
           indicator_loss, global_step=global_step, name='optimizer_shape')
 
-    return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)
+    return tf_estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)
 
   return _model_fn
 

tensorflow_graphics/projects/nasa/train.py

@@ -15,6 +15,7 @@
 import numpy as np
 import tensorflow.compat.v1 as tf
 
+from tensorflow.compat.v1 import estimator as tf_estimator
 from tensorflow_graphics.projects.nasa.lib import datasets
 from tensorflow_graphics.projects.nasa.lib import models
 from tensorflow_graphics.projects.nasa.lib import utils
@@ -45,13 +46,13 @@ def main(unused_argv):
 
   # Set up training.
   logging.info("=> Setting up training ...")
-  run_config = tf.estimator.RunConfig(
+  run_config = tf_estimator.RunConfig(
       model_dir=FLAGS.train_dir,
       save_checkpoints_steps=FLAGS.save_every,
       save_summary_steps=FLAGS.summary_every,
       keep_checkpoint_max=None,
   )
-  trainer = tf.estimator.Estimator(
+  trainer = tf_estimator.Estimator(
       model_fn=model_fn,
       config=run_config,
   )