Segmentation Fault on Deleting Detected Objects from the Buffer

[anshola]

DeepStream Version - 7.0
Docker Image - nvcr.io/nvidia/deepstream:7.0-gc-triton-devel
GPU - RTX A6000
NVIDIA GPU Driver - 535.183.01

This is a simplified version of the code I’m running inside a source probe function attached to the detection plugin in the pipeline -

while l_obj is not None:
    try:
        obj_meta = pyds.NvDsObjectMeta.cast(l_obj.data)
        if obj_meta is not None:
            if """some condition is true""":
                pyds.nvds_remove_obj_meta_from_frame(frame_meta, obj_meta)
    except StopIteration:
        break

    try:
        l_obj = l_obj.next
    except StopIteration:
        break

This is throwing segmentation faults sometimes.
Does removing an obj_meta also delete the reference to l_obj due to which l_obj.next becomes non-existent?
Do I need to accumulate all the object references I have to delete and remove them together after the object loop?

[junshengy]

@anshola

Does removing an obj_meta also delete the reference to l_obj due to which l_obj.next becomes non-existent?
Do I need to accumulate all the object references I have to delete and remove them together after the object loop?

Yes, you can’t iterate and delete a linked list at the same time

Here is the correct way to do it. I deleted all person categories.

diff --git a/apps/deepstream-test1/deepstream_test_1.py b/apps/deepstream-test1/deepstream_test_1.py
index 1367fb4..37ed2a2 100755
--- a/apps/deepstream-test1/deepstream_test_1.py
+++ b/apps/deepstream-test1/deepstream_test_1.py
@@ -43,6 +43,7 @@ def osd_sink_pad_buffer_probe(pad,info,u_data):
         print("Unable to get GstBuffer ")
         return
 
+    person_objs = []
     # Retrieve batch metadata from the gst_buffer
     # Note that pyds.gst_buffer_get_nvds_batch_meta() expects the
     # C address of gst_buffer as input, which is obtained with hash(gst_buffer)
@@ -75,13 +76,19 @@ def osd_sink_pad_buffer_probe(pad,info,u_data):
                 obj_meta=pyds.NvDsObjectMeta.cast(l_obj.data)
             except StopIteration:
                 break
-            obj_counter[obj_meta.class_id] += 1
-            obj_meta.rect_params.border_color.set(0.0, 0.0, 1.0, 0.8) #0.8 is alpha (opacity)
+            if obj_meta.class_id == PGIE_CLASS_ID_PERSON:
+                person_objs.append(obj_meta)
+            else:
+                obj_counter[obj_meta.class_id] += 1
+                obj_meta.rect_params.border_color.set(0.0, 0.0, 1.0, 0.8) #0.8 is alpha (opacity)
             try: 
                 l_obj=l_obj.next
             except StopIteration:
                 break
 
+        for person in person_objs:
+            pyds.nvds_remove_obj_meta_from_frame(frame_meta, person)
+
         # Acquiring a display meta object. The memory ownership remains in
         # the C code so downstream plugins can still access it. Otherwise
         # the garbage collector will claim it when this probe function exits.
@@ -116,7 +123,7 @@ def osd_sink_pad_buffer_probe(pad,info,u_data):
             l_frame=l_frame.next
         except StopIteration:
             break
-                       
+
     return Gst.PadProbeReturn.OK  

[anshola]

Thanks so much, this solves the issue!!!