onnx模型转入rknn3399平台上工作记录
1.rknn虚拟环境使用时报错问题
使用rknn17环境的报错:
ImportError: libdc1394.so.22: cannot open shared object file: No such file or directory
参考链接:https://blog.csdn.net/2301_80032564/article/details/142316410
创作软连接:
sudo ln -s /usr/lib/x86_64-linux-gnu/libdc1394.so.25 /usr/lib/x86_64-linux-gnu/libdc1394.so.22
ImportError: libavcodec.so.58: cannot open shared object file: No such file or directory
sudo apt install ffmpeg
/home/fuxueping/fxp/git/2/whc30001_ocr_card_recognition/model/card_detection_single/20250508/last_rknn_rotate.onnx
启动语句:python3 -m rknn.bin.visualization
2.转主干是mobilenetV3时的报错
问题分析参考:tf.nn.depthwise_conv2d()使用_depthwiseconv2d-CSDN博客
然后通过Netron查看网络结构,发现mobilenetV3中stride中有不等长的情况,具体的网络结构如下:
上图中蓝色椭圆形框中的stride为2,
解决办法:然后我将其强行改成了2
3.转yolov8检测模型时的报错
2025-05-23 15:58:10.642092: W tensorflow/core/framework/allocator.cc:107] Allocation of 1105920000 exceeds 10% of system memory.
解决办法:将量化的batchsize改小一点