diff --git a/rslp/landsat_vessels/predict_pipeline.py b/rslp/landsat_vessels/predict_pipeline.py
index 848ab70d..99de4dd2 100644
--- a/rslp/landsat_vessels/predict_pipeline.py
+++ b/rslp/landsat_vessels/predict_pipeline.py
@@ -210,8 +210,8 @@ def predict_pipeline(
 
     # Use temporary directory if scratch_path or crop_path are not specified.
     if scratch_path is None:
-        with tempfile.TemporaryDirectory() as tmp_dir:
-            scratch_path = tmp_dir
+        tmp_dir = tempfile.TemporaryDirectory()
+        scratch_path = tmp_dir.name
     else:
         tmp_dir = None
 
@@ -296,6 +296,20 @@ def predict_pipeline(
 
     json_data = []
     for idx, detection in enumerate(detections):
+        # Get longitude/latitude.
+        src_geom = STGeometry(
+            detection.projection, shapely.Point(detection.col, detection.row), None
+        )
+        dst_geom = src_geom.to_projection(WGS84_PROJECTION)
+        lon = dst_geom.shp.x
+        lat = dst_geom.shp.y
+
+        json_dict = dict(
+            longitude=lon,
+            latitude=lat,
+            score=detection.score,
+        )
+
         # Load crops from the window directory.
         images = {}
         for band in ["B2", "B3", "B4", "B8"]:
@@ -333,23 +347,10 @@ def predict_pipeline(
             with b8_fname.open("wb") as f:
                 Image.fromarray(images["B8"]).save(f, format="PNG")
 
-        # Get longitude/latitude.
-        src_geom = STGeometry(
-            detection.projection, shapely.Point(detection.col, detection.row), None
-        )
-        dst_geom = src_geom.to_projection(WGS84_PROJECTION)
-        lon = dst_geom.shp.x
-        lat = dst_geom.shp.y
+            json_dict["rgb_fname"] = str(rgb_fname)
+            json_dict["b8_fname"] = str(b8_fname)
 
-        json_data.append(
-            dict(
-                longitude=lon,
-                latitude=lat,
-                score=detection.score,
-                rgb_fname=str(rgb_fname),
-                b8_fname=str(b8_fname),
-            )
-        )
+        json_data.append(json_dict)
 
     time_profile["write_json_and_crops"] = time.time() - step_start_time