使用caffe版本的retinaface转换的retinaface, khadasvim3开发板运行错误

[lovehuanhuan]

sudo] password for khadas:
[ WARN:0] global ../modules/videoio/src/cap_gstreamer.cpp (935) open OpenCV | GStreamer warning: Cannot query video position: status=0, value=-1, duration=-1
authorise ok
tengine-lite library version: 1.4-dev
dims 1440 3840
W [vsi_nn_SortGraphNode:1278]Unprocessed node 2
W [vsi_nn_SetupGraph:621]Sort graph nodes failure.
Tengine Fatal: Pre-run subgraph(2) on TIMVX failed.
Tengine: Scheduler(sync) prerun failed.
Prerun graph failed

屏蔽prerun_graph后，程序运行后，run_graph ok，但是get_graph_output_tensor get_tensor_shape返回的out_dim异常，get_tensor_quant_param get_tensor_buffer_size返回的数量为1，无法取出返回值。模型图片如下：我认为开发板npu需要有固定尺寸输入的模型，但是output的shape好像都不在了，是不是模型转化出了问题
下面是有问题的转换后的模型

以前我用rk3399 retinaface rknn，工作正常。转换后的模型如下：