From 98036f80c6f84664a92a57949ae501d9559c531e Mon Sep 17 00:00:00 2001
From: Guillaume Caya <guillaume.caya.letourneau@gmail.com>
Date: Sat, 7 Dec 2019 18:27:34 -0500
Subject: [PATCH] Use of Adam optimizer for segmentation

---
 src/train.py | 3 +--
 src/utils.py | 2 --
 2 files changed, 1 insertion(+), 4 deletions(-)

diff --git a/src/train.py b/src/train.py
index e1ca85f..d28688a 100644
--- a/src/train.py
+++ b/src/train.py
@@ -137,8 +137,7 @@ def train():
         )
         model.compile(
             loss=jaccard_distance,
-            # loss=dice_coef_multilabel_builder(args.class_count),
-            optimizer=RMSprop(lr=args.learning_rate, momentum=0.9),
+            optimizer=Adam(lr=args.learning_rate),
             metrics=["accuracy"],
         )
     elif args.task == 'classification':
diff --git a/src/utils.py b/src/utils.py
index 05de20b..97bdd17 100644
--- a/src/utils.py
+++ b/src/utils.py
@@ -46,5 +46,3 @@ def parse_labelme_file(sample_path):
             mask[fill_row_coords, fill_col_coords] = 255
             masks.append((int(shape["label"]), mask))
     return arr, masks
-
-