rk3568 E RKNN: output dtype is undefine!

[Lahahaha]

问题描述

运行结果如下：rknn模型加载正常，推理结果正常，但每次推理都输出 output dtype is undefine! 不胜其烦！！！

检查模型文件头: 0x52 0x4b 0x4e 0x4e 0x00 0x00 0x00 0x00
正在加载RKNN模型: bisnet_mono_rk3568_2.rknn (大小: 5851141 字节)
Model loaded: input num: 1, output num: 1
Input tensors:
  index=0, name=x, n_dims=4, dims=[1, 360, 240, 1], n_elems=86400, size=86400, fmt=NHWC, type=INT8, qnt_type=AFFINE, zp=-2, scale=0.007721
Output tensors:
  index=0, name=argmax_0.tmp_0, n_dims=3, dims=[1, 360, 240, 0], n_elems=86400, size=345600, fmt=UNDEFINED, type=INT32, qnt_type=AFFINE, zp=0, scale=1.000000
RKNN model loaded successfully
E RKNN: [17:18:25.673] output dtype is undefine!
E RKNN: [17:18:25.711] output dtype is undefine!
E RKNN: [17:18:25.749] output dtype is undefine!
E RKNN: [17:18:25.786] output dtype is undefine!
E RKNN: [17:18:25.824] output dtype is undefine!
E RKNN: [17:18:25.862] output dtype is undefine!
E RKNN: [17:18:25.900] output dtype is undefine!
E RKNN: [17:18:25.937] output dtype is undefine!
E RKNN: [17:18:25.975] output dtype is undefine!
E RKNN: [17:18:26.013] output dtype is undefine!
E RKNN: [17:18:26.051] output dtype is undefine!
E RKNN: [17:18:26.088] output dtype is undefine!
E RKNN: [17:18:26.126] output dtype is undefine!
E RKNN: [17:18:26.164] output dtype is undefine!
E RKNN: [17:18:26.202] output dtype is undefine!
E RKNN: [17:18:26.240] output dtype is undefine!

额外信息

rknn 模型由 onnx 转换而来，转换配置如下：

    target = 'rk3568'

    # Create RKNN object
    rknn = RKNN()

    # Create output directory if it doesn't exist
    output_dir = os.path.dirname(RKNN_MODEL)
    if not os.path.exists(output_dir):
        os.makedirs(output_dir)

    print('--> Config model')
    rknn.config(mean_values=[[128]],
                std_values=[[128]],
                quant_img_RGB2BGR=False,
                target_platform=target,
                output_optimize=True)
    print('done')

模型转换的输出：

I rknn-toolkit2 version: 2.3.2
--> Config model
done
--> Loading model
done
--> Building model
I Loading : 100%|█████████████████████████████████████████████| 348/348 [00:00<00:00, 165059.12it/s]
I OpFusing 0: 100%|█████████████████████████████████████████████| 100/100 [00:00<00:00, 1162.50it/s]
I OpFusing 1 : 100%|█████████████████████████████████████████████| 100/100 [00:00<00:00, 392.27it/s]
I OpFusing 0 : 100%|█████████████████████████████████████████████| 100/100 [00:00<00:00, 370.18it/s]
I OpFusing 1 : 100%|█████████████████████████████████████████████| 100/100 [00:00<00:00, 351.45it/s]
I OpFusing 2 : 100%|█████████████████████████████████████████████| 100/100 [00:00<00:00, 231.14it/s]
I GraphPreparing : 100%|███████████████████████████████████████| 111/111 [00:00<00:00, 14402.27it/s]
I Quantizating 1/3: 100%|█████████████████████████████████████████| 111/111 [00:07<00:00, 14.32it/s]
I Quantizating 2/3: 100%|█████████████████████████████████████████| 111/111 [00:07<00:00, 14.62it/s]
I Quantizating 3/3: 100%|█████████████████████████████████████████| 111/111 [00:05<00:00, 20.71it/s]
W build: The default input dtype of 'x' is changed from 'float32' to 'int8' in rknn model for performance!
                       Please take care of this change when deploy rknn model with Runtime API!
I rknn building ...
E RKNN: [15:54:19.167] Unkown op target: 0
E RKNN: [15:54:19.167] Unkown op target: 0
E RKNN: [15:54:19.170] Unkown op target: 0
I rknn building done.
done
--> Export RKNN model: assets/bisnet_mono_rk3568_2.rknn
done