diff --git a/model.py b/model.py
index 302e111..fb91c65 100644
--- a/model.py
+++ b/model.py
@@ -783,7 +783,10 @@ def refine_detections(rois, probs, deltas, window, config):
     # Filter out low confidence boxes
     if config.DETECTION_MIN_CONFIDENCE:
         keep_bool = keep_bool & (class_scores >= config.DETECTION_MIN_CONFIDENCE)
-    keep = torch.nonzero(keep_bool)[:,0]
+    detections = torch.nonzero(keep_bool)
+    if not len(detections):
+        return
+    keep = detections[:,0]
 
     # Apply per-class NMS
     pre_nms_class_ids = class_ids[keep.data]
@@ -1596,7 +1599,8 @@ def detect(self, images):
 
         # Run object detection
         detections, mrcnn_mask = self.predict([molded_images, image_metas], mode='inference')
-
+        if detections is None:
+            return
         # Convert to numpy
         detections = detections.data.cpu().numpy()
         mrcnn_mask = mrcnn_mask.permute(0, 1, 3, 4, 2).data.cpu().numpy()
@@ -1671,7 +1675,8 @@ def set_bn_eval(m):
             # Detections
             # output is [batch, num_detections, (y1, x1, y2, x2, class_id, score)] in image coordinates
             detections = detection_layer(self.config, rpn_rois, mrcnn_class, mrcnn_bbox, image_metas)
-
+            if detections is None:
+                return None, None
             # Convert boxes to normalized coordinates
             # TODO: let DetectionLayer return normalized coordinates to avoid
             #       unnecessary conversions