search

PaddlePaddle / PaddleDetection

Object Detection toolkit based on PaddlePaddle. It supports object detection, instance segmentation, multiple object tracking and real-time multi-person keypoint detection.
Apache License 2.0
12.82k stars 2.89k forks source link

PP-YOLO - performance issue #3256

Closed k-radlak closed 3 years ago

k-radlak commented 3 years ago

Hi, we have tested the pretreined model on COCO dataset and we received only ~11 FPS using Tesla V100. We used the pretrained COCO model, which performance is reported as 72.9 FPS. We used this model:

image

Could you give us any hints how we can reproduce your performance? We called the inference script with the default parameters that you provided.

k-radlak commented 3 years ago

Ok. I found the solution. The model needs to be run using https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.1/deploy and TensorRT