onnx转rv1126对应的rknn时，转置卷积的输出形状会变小

[sixcen]

onnx转rv1126对应的rknn时，转置卷积的输出形状会变小，根据日志是由于onnx中的ConvTranspose在转换时会被分为DECONVOLUTION和CONV2D，而在DECONVOLUTION后形状就会变小，这个问题应该怎样解决？

--> Loading model
D Process onnx_2 ...
D RKNN output shape(input): (1 8 8 192)
D Tensor @onnx_2:out0 type: float32
D Process ConvTranspose_/layer/ConvTranspose_1 ...
D RKNN output shape(deconvolution): (1 16 16 288)
D Tensor @ConvTranspose_/layer/ConvTranspose_1:out0 type: float32
D Process attach_ConvTranspose_/layer/ConvTranspose/out0_0 ...
D RKNN output shape(output): (1 16 16 288)
D Tensor @attach_ConvTranspose_/layer/ConvTranspose/out0_0:out0 type: float32
I Build main_graph complete.

--> Building model
D Process onnx_2 ...
D RKNN output shape(input): (1 8 8 192)
D Tensor @onnx_2:out0 type: float32
D Process ConvTranspose_/layer/ConvTranspose_1_plane_3 ...
D RKNN output shape(deconvolution): (1 15 15 192)
D Tensor @ConvTranspose_/layer/ConvTranspose_1_plane_3:out0 type: float32
D Process ConvTranspose_/layer/ConvTranspose_1_dual_conv_4 ...
D RKNN output shape(convolution): (1 15 15 288)
D Tensor @ConvTranspose_/layer/ConvTranspose_1_dual_conv_4:out0 type: float32
D Process attach_ConvTranspose_/layer/ConvTranspose/out0_0 ...
D RKNN output shape(output): (1 15 15 288)
D Tensor @attach_ConvTranspose_/layer/ConvTranspose/out0_0:out0 type: float32
I Build main_graph complete.