There are Re-ID index in each object of each source. It is fine for multiple streams.
Hi, thank you for your response.
Where is that index? Since we are writing the binding ourselves.
I actually can’t even access the tensor in this way:
if user_meta and user_meta.base_meta.meta_type == pyds.NVDS_TRACKER_OBJ_REID_META and user_meta.user_meta_data:
tensor_meta = pyds.NvDsInferTensorMeta.cast(user_meta.user_meta_data)
layer = pyds.get_nvds_LayerInfo(tensor_meta, 0)
ptr = ctypes.cast(pyds.get_ptr(layer.buffer),
ctypes.POINTER(ctypes.c_float))
features = np.ctypeslib.as_array(ptr, shape=(512,))
print(features.shape)
Am I doing it correct?
Please use these updated bindings for tracker meta while compiling deepstream python apps.
bindtrackermeta.cpp.txt (6.7 KB)
trackermetadoc.h.txt (10.5 KB)
nvdsmetadoc.h.txt (23.7 KB)
bindnvdsmeta.cpp.txt (27.5 KB)
Example Usage:
::
gst_buffer = info.get_buffer()
if not gst_buffer:
print("Unable to get GstBuffer ")
return
batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))
l_batch_user=batch_meta.batch_user_meta_list #
while l_batch_user is not None:
try:
user_meta= pyds.NvDsUserMeta.cast(l_batch_user.data)
except StopIteration:
break
if user_meta and user_meta.base_meta.meta_type == pyds.NVDS_TRACKER_BATCH_REID_META:
pReidTensor = pyds.NvDsReidTensorBatch.cast(user_meta.user_meta_data)
try:
l_batch_user=l_batch_user.next
except StopIteration:
break
l_frame = batch_meta.frame_meta_list
while l_frame is not None:
try:
frame_meta = pyds.NvDsFrameMeta.cast(l_frame.data)
except StopIteration:
break
if not pReidTensor:
continue
l_obj = frame_meta.obj_meta_list
while l_obj is not None:
try:
obj = pyds.NvDsObjectMeta.cast(l_obj.data)
except StopIteration:
break
id = obj.object_id
l_obj_user = obj.obj_user_meta_list
obj_user_meta = pyds.NvDsUserMeta.cast(l_obj_user.data)
if obj_user_meta and obj_user_meta.base_meta.meta_type == pyds.NVDS_TRACKER_OBJ_REID_META and obj_user_meta.user_meta_data:
reidInd_ptr = ctypes.cast(pyds.get_ptr(obj_user_meta.user_meta_data), ctypes.POINTER(ctypes.c_int32))
reidInd = reidInd_ptr.contents.value
# Check the conditions
if 0 <= reidInd < pReidTensor.numFilled:
features = pReidTensor.ptr_host[reidInd]
features[id] = features.tolist() #This will be the Re-ID feature for the obj. id
try:
l_obj = l_obj.next
except StopIteration:
break
This is the example usage code you can use to get reid feature. @JoeShz
@kesong Please help update the repo if you find the bindingd are correct. Will be useful for other developers in future.
Regards
Thanks for your sharing. We will review it.
I built the binding using Cmake. It was built successfully. But, I get this error:
if user_meta and user_meta.base_meta.meta_type == pyds.NVDS_TRACKER_BATCH_REID_META:
AttributeError: module 'pyds' has no attribute 'NVDS_TRACKER_BATCH_REID_META'
I am sure that there’s another file to be edited since It was working before xD but I don’t remember which file I had to edit.
@st123439
1 Like
copy the files:
.cpp files to: opt/nvidia/deepstream/deepstream-6.3/sources/deepstream_python_apps/bindings/src/
.h files to : opt/nvidia/deepstream/deepstream-6.3/sources/deepstream_python_apps/bindings/docstrings/
Then :
cd /opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/bindings && \
rm -rf build && \
mkdir build && \
cd build && \
cmake .. -DPYTHON_MAJOR_VERSION=3 -DPYTHON_MINOR_VERSION=8 -DPIP_PLATFORM=linux_x86_64 -DDS_PATH=/opt/nvidia/deepstream/deepstream/ && \
make -j$(nproc) && \
echo "###########################" && \
echo "Installing built PyDS wheel" && \
echo "###########################" && \
pip3 install ./pyds-1.1.8-py3-none*.whl
This is how you can compile the bindings.
Best luck!
1 Like
Thanks a lot for sharing binding related files. I followed your instruction and I used sample code you provided. However, when I run the code, it says: “if not pReidTensor:
UnboundLocalError: local variable ‘pReidTensor’ referenced before assignment”
So I tried to declare pReidTensor, but it just freezes.
def pgie_src_pad_buffer_probe(pad, info, u_data):
frame_number = 0
pReidTensor = None
num_rects = 0
gst_buffer = info.get_buffer()
if not gst_buffer:
print("Unable to get GstBuffer ")
return
batch_meta = pyds.gst_buffer_get_nvds_batch_meta(hash(gst_buffer))
l_batch_user=batch_meta.batch_user_meta_list #
while l_batch_user is not None:
try:
user_meta= pyds.NvDsUserMeta.cast(l_batch_user.data)
except StopIteration:
break
if user_meta and user_meta.base_meta.meta_type == pyds.NVDS_TRACKER_BATCH_REID_META:
pReidTensor = pyds.NvDsReidTensorBatch.cast(user_meta.user_meta_data)
try:
l_batch_user=l_batch_user.next
except StopIteration:
break
l_frame = batch_meta.frame_meta_list
while l_frame is not None:
try:
frame_meta = pyds.NvDsFrameMeta.cast(l_frame.data)
except StopIteration:
break
if pReidTensor is None:
# if not pReidTensor:
continue
l_obj = frame_meta.obj_meta_list
while l_obj is not None:
try:
obj = pyds.NvDsObjectMeta.cast(l_obj.data)
except StopIteration:
break
id = obj.object_id
l_obj_user = obj.obj_user_meta_list
obj_user_meta = pyds.NvDsUserMeta.cast(l_obj_user.data)
if obj_user_meta and obj_user_meta.base_meta.meta_type == pyds.NVDS_TRACKER_OBJ_REID_META and obj_user_meta.user_meta_data:
reidInd_ptr = ctypes.cast(pyds.get_ptr(obj_user_meta.user_meta_data), ctypes.POINTER(ctypes.c_int32))
reidInd = reidInd_ptr.contents.value
# Check the conditions
if 0 <= reidInd < pReidTensor.numFilled:
features = pReidTensor.ptr_host[reidInd]
features[id] = features.tolist() #This will be the Re-ID feature for the obj. id
try:
l_obj = l_obj.next
except StopIteration:
break
return Gst.PadProbeReturn.OK
Warning: gst-stream-error-quark: No decoder available for type 'application/x-rtp, media=(string)application, payload=(int)107, clock-rate=(int)90000, encoding-name=(string)VND.ONVIF.METADATA, a-tool=(string)"LIVE555\ Streaming\ Media\ v2022.01.21", a-type=(string)broadcast, x-qt-text-nam=(string)"RTSP/RTP\ stream\ from\ Encoder", x-qt-text-inf=(string)cam0_0, npt-start=(guint64)0, play-speed=(double)1, play-scale=(double)1, ssrc=(uint)2647244233'. (6): gsturidecodebin.c(920): unknown_type_cb (): /GstPipeline:pipeline0/GstBin:source-bin-01/GstURIDecodeBin:uri-decode-bin
Decodebin child added: h264parse1
Decodebin child added: capsfilter1
Decodebin child added: nvv4l2decoder1
In cb_newpad
gstname= video/x-raw
features= <Gst.CapsFeatures object at 0x7f39d0f8f580 (GstCapsFeatures at 0x7f38b002b460)>
Best Regards
I think its because of the probe. I see that the probe function you have used is after the pgie. But we need to use this probe after the tracker if i understand the issue correctly in your case.
And in case of the freezing, i think its because you have not iterated the frame as :
try:
l_frame = l_frame.next
except StopIteration:
break
So once you go inside the probe, the logic was able to go in the first frame and couldnt iterate over the other frames.
1 Like
Also, make sure that outputReidTensor equals 1 in your tracker.
Thanks for your reply.
Using probe after the track solved the problem.
Now I can access to Re-ID features.
I have a new question regarding the code below
# Check the conditions
if 0 <= reidInd < pReidTensor.numFilled:
features = pReidTensor.ptr_host[reidInd]
features[id] = features.tolist() #This will be the Re-ID feature for the obj. id
I don’t understand why you added the last part.
Is there any specific reason you changed features to list? And is the type of features Dictionary?
In addition, if possible, could you provide me any insights to track objects with those features on different camera (with different view, not at the same time.)
Best Regards,
The features are the list of array. So i want to save it as an array.
the id is the obj id for which the features are calculated.
So if you want to do reid in another camera, you need to save these features in the buffer and reuse it. Depends on what you want to do.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.