Replies: 5 comments
-
可以详细说明一下问题吗? |
Beta Was this translation helpful? Give feedback.
-
PaddleOCRv4 server端文字检测模型转ONNX准确率并未下降,ONNX转TRT后精度下降70%左右,可以参照之前有几个开发者提出了同样问题: |
Beta Was this translation helpful? Give feedback.
-
能否帮忙解答一下?多谢 |
Beta Was this translation helpful? Give feedback.
-
可以尝试使用Fastdeploy进行推理,切换后端为trt:
参考:https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/ocr/PP-OCR/cpu-gpu/python 但需要注意,目前没有在FP16上做过精度验证,确实可能存在精度损失的情况,建议先使用FP32。 |
Beta Was this translation helpful? Give feedback.
-
谢谢您的答复,检测模型转tensorRT,FP32的看起来也是有问题的,您那边能验证FP32是无问题的? |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
ppocrv4 检测模型转tensorrt性能下降严重,能否给出解决方案?多谢
Beta Was this translation helpful? Give feedback.
All reactions