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
12.77k stars 2.88k forks source link

yolo模型通过离线量化模型转换为Int8后推理报错! #7412

Open lilianjie111111 opened 1 year ago

lilianjie111111 commented 1 year ago

问题确认 Search before asking

请提出你的问题 Please ask your question

image image

lyuwenyu commented 1 year ago

具体是那个模型啊 直接导出的嘛

yghstill commented 1 year ago

@lilianjie111111 quant_post_dynamic是动态离线量化,对模型没有任何加速,并且只能在Paddle Lite上部署。推荐直接使用静态离线量化,支持的更好。https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/api_cn/static/quant/quantization_api.rst#quant_post_static demo可参考:https://github.com/PaddlePaddle/PaddleSlim/tree/develop/example/post_training_quantization

lilianjie111111 commented 1 year ago

具体是那个模型啊 直接导出的嘛 yoloe

lilianjie111111 commented 1 year ago

@lilianjie111111 quant_post_dynamic是动态离线量化,对模型没有任何加速,并且只能在Paddle Lite上部署。推荐直接使用静态离线量化,支持的更好。https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/api_cn/static/quant/quantization_api.rst#quant_post_static demo可参考:https://github.com/PaddlePaddle/PaddleSlim/tree/develop/example/post_training_quantization

这边尝试了通用检测python端的推理也没有发现有明显的加速呀,甚至更慢 ![Uploading image.png…]()