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
1.39k stars 345 forks source link

Jetson poor performance in YOLOv5 and YOLOR models #138

Closed marcoslucianops closed 2 years ago

marcoslucianops commented 2 years ago

Found poor performance in YOLOv5 and YOLOR models in Jetson platform. Working in a fix for it.

Issues:

marcoslucianops commented 2 years ago

Repo updated, please use the new files.

Results

4x faster inference in AGX using YOLOv5n model in FP16 mode

CPU YOLO Decoder GPU YOLO Decoder 
PERF:  50.57 (50.33)
PERF:  50.27 (50.32)
PERF:  50.30 (50.30)
PERF:  50.76 (50.33)
**PERF:  50.80 (50.35)
 
PERF:  223.94 (216.90)
PERF:  224.58 (219.91)
PERF:  223.49 (221.00)
PERF:  224.36 (221.60)
**PERF:  224.39 (222.16)