deploy RK3588 rknn model in runtime but get infer result as all zero

[ken4647]

.rknn model is transform from rknn2 and it can run correctly in rktoolkit2 in PC.
However, it got wrong when using 'runtime' to infer and got all zero result, while the shape of result is correct. The input has already checked.(there is no example in runtime usage, while my model is not used as any way in 'vision' example).

The code is below:

import fastdeploy as fd
from fastdeploy import ModelFormat
import numpy as np
import cv2

camera = cv2.VideoCapture(2)
option = fd.RuntimeOption()

option.set_model_path("mymodel.rknn", model_format=ModelFormat.RKNN)

option.use_cpu()
option.use_rknpu2(rknpu2_name=fd.c_lib_wrap.CpuName.RK3588,rknpu2_core=fd.c_lib_wrap.CoreMask.RKNN_NPU_CORE_AUTO)

runtime = fd.Runtime(option)

input_name = runtime.get_input_info(0).name

while True:
    ret,frame = camera.read()
    cv2.imshow("test",frame)
    cv2.waitKey(1)
    rgb_array = cv2.resize(frame,(320,176))
    hsv_array = cv2.cvtColor(rgb_array,cv2.COLOR_BGR2HSV).reshape((1,176,320,3))
    # print(hsv_array)
    results = runtime.infer({
        input_name: hsv_array
    })
    input_check = hsv_array
    output_check = runtime.num_outputs()
    print(input_check)
    print(output_check)
    print(results)

[ken4647]

OK,the problem is at model transformation form .pt to .rknn.I fixed it by transform .pt to onnx then to .rknn.Now it works!