[RKNPU2] Update quantitative model (#879)

* 对RKNPU2后端进行修改,当模型为非量化模型时,不在NPU执行normalize操作,当模型为量化模型时,在NUP上执行normalize操作

* 更新RKNPU2框架,输出数据的数据类型统一返回fp32类型

* 更新scrfd,拆分disable_normalize和disable_permute

* 更新scrfd代码,支持量化

* 更新scrfd python example代码

* 更新模型转换代码,支持量化模型

* 更新文档

* 按照要求修改

* 按照要求修改

* 修正模型转换文档

* 更新一下转换脚本
This commit is contained in:
Zheng_Bicheng
2022-12-19 13:58:43 +08:00
committed by GitHub
parent 383887dede
commit dc13eb7049
18 changed files with 92 additions and 148 deletions

View File

@@ -51,11 +51,17 @@ class SCRFD(FastDeployModel):
return self._model.predict(input_image, conf_threshold,
nms_iou_threshold)
def disable_normalize_and_permute(self):
def disable_normalize(self):
"""
This function will disable normalize and hwc2chw in preprocessing step.
This function will disable normalize in preprocessing step.
"""
self._model.disable_normalize_and_permute()
self._model.disable_normalize()
def disable_permute(self):
"""
This function will disable hwc2chw in preprocessing step.
"""
self._model.disable_permute()
# 一些跟SCRFD模型有关的属性封装
# 多数是预处理相关可通过修改如model.size = [640, 640]改变预处理时resize的大小前提是模型支持