用户您好,请详细描述您所遇到的问题,这会帮助我们快速定位问题~
1.芯片型号:X3派
2.天工开物开发包OpenExplorer版本:hrt_model_exec -v : 1.18.2
3.问题定位:
1. 执行 04_inference.sh origin 结果正常,执行 04_inference.sh quanti结果异常,输出为None. 详情如下:
2. 使用bin文件模型进行推理,报错如下:
hrt_model_exec model_info --model_file models/picodet_xs_320x320_rgb.binI0000 00:00:00.000000 15620 vlog_is_on.cc:197] RAW: Set VLOG level for "*" to 3[BPU_PLAT]BPU Platform Version(1.3.3)![HBRT] set log level as 0. version = 3.15.25.0[DNN] Runtime version = 1.18.2_(3.15.25 HBRT)[A][DNN][packed_model.cpp:234][Model](2023-09-21,14:14:36.135.346) [HorizonRT] The model builder version = 1.18.2Load model to DDR cost 297.965ms.This model file has 1 model:[picodet_xs_320x320_rgb]---------------------------------------------------------------------[model name]: picodet_xs_320x320_rgbinput[0]: name: imageinput source: HB_DNN_INPUT_FROM_PYRAMIDvalid shape: (1,3,320,320,)aligned shape: (1,3,320,320,)aligned byte size: 153600tensor type: HB_DNN_IMG_TYPE_NV12tensor layout: HB_DNN_LAYOUT_NCHWquanti type: NONEstride: (0,0,0,0,)output[0]: name: p2o.Mul.179valid shape: (1,2125,4,1,)aligned shape: (1,2125,4,1,)aligned byte size: 34000tensor type: HB_DNN_TENSOR_TYPE_F32tensor layout: HB_DNN_LAYOUT_NCHWquanti type: NONEstride: (34000,16,4,4,)output[1]: name: p2o.Concat.9valid shape: (1,1,2125,1,)aligned shape: (1,1,2125,1,)aligned byte size: 8500tensor type: HB_DNN_TENSOR_TYPE_F32tensor layout: HB_DNN_LAYOUT_NCHWquanti type: NONEstride: (8500,8500,4,4,)---------------------------------------------------------------------
4.问题具体描述:
执行03_build.sh模型转换脚本,origin模型推理正常,quanti量化模型掉点严重,hb_mapper_makertbin.log日志详情可见附件,但不清楚哪里出现问题,改如何解决?
附:
模型转换配置及相关模型文件:链接:https://pan.baidu.com/s/1k1GIwIuGvI0IAzRU9kkLfQ?pwd=8xay 提取码:8xay
hb_mapper_makertbin.log-
inference_origin.log-
inference_quanti.log