search

marcoslucianops / DeepStream-Yolo

NVIDIA DeepStream SDK 7.0 / 6.4 / 6.3 / 6.2 / 6.1.1 / 6.1 / 6.0.1 / 6.0 / 5.1 implementation for YOLO models
MIT License
1.45k stars 356 forks source link

Performance gap between Deepstream YOLO and PyTorch YOLO #494

Open jatinjotwani opened 10 months ago

jatinjotwani commented 10 months ago

I trained a YOLOv5s model on a custom dataset (VisDrone), achieving excellent results in PyTorch. However, when I converted the model to ONNX for deployment on my Jetson Nano using DeepStream 6.0, I encountered significantly poor performance on the same image. I used the parameters "--simplify --batch 1" during conversion. Attached are the images for reference. Seeking insights on the performance gap and any optimization suggestions. Appreciate your assistance. pexels-kelly-2402233 image_50447873

maximereder commented 10 months ago

Hi @jatinjotwani I am also trying to put a model to Jetson Nano. However, I cannot run YOLO on GPU, can you tell me if you achieve this ? Thanks, Maxime

agarwalkunal12 commented 9 months ago

Experiencing the same issue of much lower accuracy after converting to ONNX using Deepstream 6.2 on the same set of images as compared to the Pytorch model. Used --dynamic flag and --batch 2 flag. Got the same accuracy issue on Yolov5 and Yolov8.