用yolov5的best.pt导出成onnx转化成fp32 bmodel后在Airbox上跑，报维度不匹配怎么处理？

用官方的模型不出错，用自己的yolov5训练出来的best.pt导出成onnx转化成fp32 bmodel后在Airbox上跑，出现报错：
linaro@bm1684:~/yolov5/python$ python3 yolov5_opencv_3output.py --bmodel ../../best_pt_1684x_int8_sym.bmodel --input ../../21.png --label=../../classes.names[BMRT][bmcpu_setup:406] INFO:cpu_lib 'libcpuop.so' is loaded.bmcpu init: skip cpu_user_definedopen usercpu.so, init user_cpu_init[BMRT][load_bmodel:1090] INFO:Loading bmodel from [../../best_pt_1684x_int8_sym.bmodel]. Thanks for your patience...[BMRT][load_bmodel:1029] INFO:pre net num: 0, load net num: 1processing file: ../../21.pngimg pre cost time 0.05923032760620117use decode data as inputinput_data shape:  (1, 3, 640, 640)(1, 25200, 11)Traceback (most recent call last):  File "yolov5_opencv_3output.py", line 461, in     result_image = yolov5.inference_center_np(frame, opt.use_np_file_as_input)  File "yolov5_opencv_3output.py", line 310, in inference_center_np    dets = self.predict_center(input_data, use_np_file_as_input)  File "yolov5_opencv_3output.py", line 201, in predict_center    bs, _, ny, nx, nc = feat.shapeValueError: not enough values to unpack (expected 5, got 3)模型有6个类，是用yolov5s模型训练出来的，转成bmodel是利用mlir转的
#进入docker容器mkdir acd    cd acddocker cp D:yolov5-masteraccidentimagestrain 8f3388b70bb3:/workspace/acddocker cp D:yolov5-masteraccidentimagesval 8f3388b70bb3:/workspace/acddocker cp D:yolov5-masterrunstrainexpmyqweightsbest.onnx 8f3388b70bb3:/workspace/acdmkdir workspace1    cd workspace1model_transform --model_name best_pt --model_def ../best.onnx --input_shapes [[1,3,640,640]] --mean 0.0,0.0,0.0 --scale 0.0039216,0.0039216,0.0039216 --keep_aspect_ratio --pixel_format rgb --test_input ../val/65.jpg --test_result best_pt_top_outputs.npz --mlir best_pt.mlirmodel_deploy --mlir best_pt.mlir --quantize F32 --processor bm1684x --test_input best_pt_in_f32.npz --test_reference best_pt_top_outputs.npz --model bestf32.bmodel跑模型是利用文件中的py文件