**• Hardware Platform (Jetson / GPU)： GPU
**• DeepStream Version： 6.2
**• TensorRT Version： 8.5
**• NVIDIA GPU Driver Version (valid for GPU only)： 535
**• Issue Type( questions, new requirements, bugs)： questions
**• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)：
I used AlexeyAB’s darknet project to train the YOLOv4 model file and then used Tianxiaomo’s pytorch-YOLOv4 project to convert the model file into a .onnx format file.
command：
$ python3 demo_darknet2onnx.py yolov4-Can.cfg yolov4-Can.names yolov4-Can.weights ./data/xxx.jpg 4
Finally, I used the trtexec command to convert the .onnx format file into a .engine file needed by DeepStream.
command：
$ trtexec --onnx=yolov4_4_3_416_416_static.onnx --explicitBatch --saveEngine=yolov4-Can.engine --workspace=4096 --fp16
In the NVIDIA-AI-IOT deepstream_python_apps project (branch nvaie-3.0), I modified the deepstream-test2 example.
PGIE is a detector, and the class is a can-shaped object, with a batch-size of 4.
SGIE1 is a classifier, and the class consists of two types of drink logo text and images, with a batch-size of 16.
(Drink_A_can, Drink_A_logo, and Drink_A_img are trained on different parts of the same object in the model.)
The program runs normally, and the Bbox only contains the detection results from PGIE and the object tracking IDs, but there are no classification results from SGIE1.
Could you please let me know if this is an issue with the model, configuration parameters, or the input/output format?
Below are my relevant configuration files:
imagedata_pgie_config.txt (3.0 KB)
imagedata_sgie1_config.txt (3.4 KB)
imagedata_tracker_config.txt (1.3 KB)
labels_Can.txt (4 Bytes)
labels_Product.txt (74 Bytes)