转换模型失败

详细错误如下，想知道是什么原因？

D Process conv2d_14_1/Relu_15 ...
D RKNN output shape(relu): (1 256 256 32)
D Process conv2d_15_1/BiasAdd_13 ...
D RKNN output shape(convolution): (1 256 256 32)
D Process conv2d_15_1/Relu_11 ...
D RKNN output shape(relu): (1 256 256 32)
D Process conv2d_transpose_3_1/BiasAdd_9 ...
D RKNN output shape(deconvolution): (1 512 512 16)
D Process concatenate_3_1/concat_8 ...
D RKNN output shape(concat): (1 512 512 32)
D Process conv2d_16_1/BiasAdd_7 ...
D RKNN output shape(convolution): (1 512 512 16)
D Process conv2d_16_1/Relu_6 ...
D RKNN output shape(relu): (1 512 512 16)
D Process conv2d_17_1/BiasAdd_5 ...
D RKNN output shape(convolution): (1 512 512 16)
D Process conv2d_17_1/Relu_4 ...
D RKNN output shape(relu): (1 512 512 16)
D Process out_1/BiasAdd_3 ...
D RKNN output shape(fullconnect): (1 512 512 3)
D Process out_1/truediv_2 ...
D RKNN output shape(softmax): (1 512 512 3)
D Process attach_out_1/truediv/out0_0 ...
D RKNN output shape(output): (1 512 512 3)
I Build Model_7_sl_detection_1625 complete.
I Config File "/Users/qliu/anaconda3/lib/python3.6/site-packages/rknn/base/RK1808_PID0X82" load/generated successfully
I Initialzing network optimizer by /Users/qliu/anaconda3/lib/python3.6/site-packages/rknn/base/RK1808_PID0X82 ...
D Optimizing network with t2c_tf2caffe
I Start T2C Switcher...
D convert concatenate_4/concat_42(concat) axis 3 to 1
D convert concatenate_1_1/concat_31(concat) axis 3 to 1
D convert concatenate_2_1/concat_19(concat) axis 3 to 1
D convert concatenate_3_1/concat_8(concat) axis 3 to 1
D insert permute conv2d_18/BiasAdd_16_RKNN_mark_perm_54 before conv2d_18/BiasAdd_16
I End T2C Switcher...
D Optimizing network with qnt_adjust_coef, multiply_transform, add_extra_io, format_input_ops, auto_fill_zero_bias, conv_kernel_transform, twod_op_transform, conv_1xn_transform, strip_op, extend_unstack_split, extend_batchnormalize, swapper, merge_layer, transform_layer, proposal_opt, broadcast_op, strip_op, auto_fill_reshape_zero, adjust_output_attrs
I insert reshape layer before softmax out_1/truediv_2
I insert reshape layer after softmax out_1/truediv_2
D Optimizing network with c2drv_convert_axis, c2drv_convert_shape, c2drv_convert_array, c2drv_cast_dtype
I Building data ...
E Catch exception when building RKNN model!
E Traceback (most recent call last):
E   File "rknn/api/rknn_base.py", line 605, in rknn.api.rknn_base.RKNNBase.build
E   File "rknn/api/rknn_base.py", line 439, in rknn.api.rknn_base.RKNNBase._build
E   File "rknn/base/ovxconfiggenerator.py", line 178, in rknn.base.ovxconfiggenerator.generate_vx_config_from_files
E   File "rknn/base/RKNNlib/app/exporter/ovxlib_case/casegenerator.py", line 452, in rknn.base.RKNNlib.app.exporter.ovxlib_case.casegenerator.CaseGenerator.generate
E   File "rknn/base/RKNNlib/app/exporter/ovxlib_case/casegenerator.py", line 177, in rknn.base.RKNNlib.app.exporter.ovxlib_case.casegenerator.CaseGenerator._build_netdict
E   File "rknn/base/RKNNlib/app/helper/striphelper.py", line 92, in rknn.base.RKNNlib.app.helper.striphelper.StripHelper.build_netdict
E   File "rknn/base/RKNNlib/app/helper/striphelper.py", line 329, in rknn.base.RKNNlib.app.helper.striphelper.StripHelper.build_data
E   File "rknn/base/RKNNlib/app/helper/transhelper.py", line 77, in rknn.base.RKNNlib.app.helper.transhelper.TransHelper.to_driver
E   File "rknn/base/RKNNlib/app/helper/transhelper.py", line 125, in rknn.base.RKNNlib.app.helper.transhelper.TransHelper.fcl_weight
E   File "/Users/qliu/anaconda3/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 292, in reshape
E     return _wrapfunc(a, 'reshape', newshape, order=order)
E   File "/Users/qliu/anaconda3/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 56, in _wrapfunc
E     return getattr(obj, method)(*args, **kwds)
E ValueError: cannot reshape array of size 48 into shape (512,512,16,3)
  7215 MainThread failed to build rknn model

E ValueError: cannot reshape array of size 48 into shape (512,512,16,3)
请问你的模型转换之前使用PC加载运行推理过么

jefferyzhang 发表于 2019-11-19 08:18
E ValueError: cannot reshape array of size 48 into shape (512,512,16,3)
请问你的模型转换之前使用PC加 ...

在 PC 上用 tensorflow 推理都是正常的

qliu 发表于 2019-11-19 10:00
在 PC 上用 tensorflow 推理都是正常的

请问：
1. 你tf是读取这个pb文件去推理正常么？
2. 读取这个pb的tf和转换rknn的tf版本是否一致
3. rknn-toolkit版本是多少

jefferyzhang 发表于 2019-11-19 10:07
请问：
1. 你tf是读取这个pb文件去推理正常么？
2. 读取这个pb的tf和转换rknn的tf版本是否一致

1. tf 读取这个 pb 文件推理是正常的
2. 用作推理的 tf 和 rknn 在一个 python 环境里，这个环境里 tf 可以推理，但是模型转换失败
3. rknn-toolkit 是 1.2.1，配的 tensorflow 是 1.11.0

帮忙把pb模型和转换脚本上传到网盘，然后发送给我们debug。

jefferyzhang 发表于 2019-11-19 11:07
帮忙把pb模型和转换脚本上传到网盘，然后发送给我们debug。

好的，我私信您了

好的，收到了

jefferyzhang 发表于 2019-11-19 14:17
好的，收到了

请问这个问题有进展吗？

哪有这么快，我只是帮忙报了bug给相关部门，他们自己有优先级排序和处理顺序，一般都得等下个版本才会发布解决的问题。
有进展我会回你帖子的