Skip to content

Latest commit

 

History

History
120 lines (113 loc) · 2.87 KB

3.5M_slim.md

File metadata and controls

120 lines (113 loc) · 2.87 KB

PPOCR 3.5M模型压缩方案

本方案使用量化和剪枝两种方法对PPOCR中模型进行压缩。

教程内容参考:https://github.com/PaddlePaddle/PaddleSlim/blob/release/2.0.0/demo/ocr/README.md AIStudio教程:https://aistudio.baidu.com/aistudio/projectdetail/898523

示例结果

序号 任务 模型 压缩策略[3][4] 精度(自建中文数据集) 耗时[1](ms) 整体耗时[2](ms) 加速比 整体模型大小(M) 压缩比例 下载链接
0 检测 MobileNetV3_DB 61.7 224 375 - 8.6 -
识别 MobileNetV3_CRNN 62.0 9.52
1 检测 SlimTextDet PACT量化训练 62.1 195 348 8% 2.8 67.82%
识别 SlimTextRec PACT量化训练 61.48 8.6
2 检测 SlimTextDet_quat_pruning 剪裁+PACT量化训练 60.86 142 288 30% 2.8 67.82%
识别 SlimTextRec PACT量化训练 61.48 8.6
3 检测 SlimTextDet_pruning 剪裁 61.57 138 295 27% 2.9 66.28%
识别 SlimTextRec PACT量化训练 61.48 8.6

注意:

  • [1] 耗时评测环境为:骁龙855芯片+PaddleLite。
  • [2] 整体耗时不等于检测耗时加识别耗时的原因是:识别模型的耗时为单个检测框的耗时,一张图片可能会有多个检测框。
  • [3] 参考下面关于OCR量化的说明
  • [4] 参考下面关于OCR剪裁的说明