How can I get a secondary-gie classifier CNN model output as object feature for reid?

[zhangquan0823]

Please provide complete information as applicable to your setup.

**• Hardware Platform (Jetson / GPU) :Jetson Xavier NX
**• DeepStream Version:Deepstream6.2
**• JetPack Version (valid for Jetson only):Jetpack5.1
**• Issue Type( questions, new requirements, bugs):questions
• How to reproduce the issue ? (This is for questions.How can I get a secondary-gie classifier CNN model output as object feature for reid?)

[miguel.taylor]

Hi @zhangquan0823

In the nvinfer element that corresponds to the SGIE you want to access, set the property output-tensor-meta=true. Then, you can retrieve the output tensor from the metadata in NvDsBatchMeta → NvDsFrameMeta → (NvDsInferTensorMeta) NvDsUserMeta.

[zhangquan0823]

Set the property output-tensor-meta=true ，I modified the callback function “all_bbox_generated” in deepstream-app to watchNvDsUserMetavariable.But the pointer to “NvDsUserMeta” is null.

the sgie property set like this:
[property]
gpu-id=0
enable-dla=1
use-dla-core=1
model-color-format=0
net-scale-factor=1.0
onnx-file=reid256.onnx
model-engine-file=reid256.onnx_b1_dla1_fp16.engine
labelfile-path=labels_reid.txt
output-tensor-meta=1
force-implicit-batch-dim=0
batch-size=1
secondary-reinfer-interval=1
process-mode=2

0=FP32, 1=INT8, 2=FP16 mode

network-mode=2
is-classifier=1
output-blob-names=732
classifier-async-mode=1
classifier-threshold=0.3
input-object-min-width=128
input-object-min-height=256
operate-on-gie-id=1
operate-on-class-ids=0
classifier-type=reid
#scaling-filter=0
#scaling-compute-hw=0

[kesong]

Please refer the sample: /opt/nvidia/deepstream/deepstream-6.3/sources/apps/sample_apps/deepstream-infer-tensor-meta-test/

[zhangquan0823]

The problem was solved, thanks！