Fixed typos (#11126)

* Fixed typos

* Update tuples_dataset.py

* Update sfm120k.py

* Update delf_model.py

Author: sushreebarsa

research/delf/delf/python/datasets/sfm120k/sfm120k.py

@@ -30,7 +30,7 @@
 def id2filename(image_id, prefix):
   """Creates a training image path out of its id name.
 
-  Used for the image mapping in the Sfm120k datset.
+  Used for the image mapping in the Sfm120k dataset.
 
   Args:
     image_id: String, image id.

research/delf/delf/python/datasets/tuples_dataset.py

@@ -198,14 +198,14 @@ def __repr__(self):
             self._num_negatives)
     fmt_str += '\tNumber of tuples processed in an epoch: {}\n'.format(
             self._num_queries)
-    fmt_str += '\tPool size for negative remining: {}\n'.format(self._pool_size)
+    fmt_str += '\tPool size for negative remaining: {}\n'.format(self._pool_size)
     return fmt_str
 
   def create_epoch_tuples(self, net):
     """Creates epoch tuples with the hard-negative re-mining.
 
     Negative examples are selected from clusters different than the cluster
-    of the query image, as the clusters are ideally non-overlaping. For
+    of the query image, as the clusters are ideally non-overlapping. For
     every query image we choose  hard-negatives, that is, non-matching images
     with the most similar descriptor. Hard-negatives depend on the current
     CNN parameters. K-nearest neighbors from all non-matching images are

research/delf/delf/python/training/model/delf_model.py

@@ -34,7 +34,7 @@ class AttentionModel(tf.keras.Model):
 
   Uses two [kernel_size x kernel_size] convolutions and softplus as activation
   to compute an attention map with the same resolution as the featuremap.
-  Features l2-normalized and aggregated using attention probabilites as weights.
+  Features l2-normalized and aggregated using attention probabilities as weights.
   The features (targets) to be aggregated can be the input featuremap, or a
   different one with the same resolution.
   """