Please provide complete information as applicable to your setup.
• Hardware Platform ( GPU : RTX A4000)
• DeepStream Version : 7.1
**• NVIDIA GPU Driver Version : 535.230 **
• Issue Type : bugs)
• 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 have developed a gender classification model based on EfficientNet. When integrating this model as the primary GIE (PGIE) within a DeepStream pipeline—following the guidelines provided in this NVIDIA forum post - https://forums.developer.nvidia.com/t/how-to-deploy-classification-application-on-deepstream/259853/19 —I observed that approximately 20% of the classification results differ from those obtained when running the same images through ONNX Runtime.
I saw a similar bug reported in the post - Classifier result on onnx doesn't match Deepstream result - #32 by Asma1_1. However the suggestions does not help at all.
For reproducing the results I am attaching the source files for both deepstream pipeline and onnxruntime
Deepstream source code - deepstream_class.zip
deepstream_class.zip (1.8 KB)
**Model inference configuration **-
config_sgi_gender_71.zip (541 Bytes)
Model file and label file -
people_attribute.zip (29.5 MB)
Onnx inference code -
onnx_inference_on_images.zip (1.3 KB)
Both source codes will infer the images and dump the classification results in csv file
**Sample images ** -
cropped_person.zip (491.1 KB)
Any help would be greatly appreciated