模型是一个官方示例附带的: github.com/PaddlePaddle/models/tree/develop/PaddleCV/face_detection

下载地址是 http://paddlemodels.bj.bcebos.com/PyramidBox_WiderFace.tar.gz

我尝试用opt工具去把下载的模型用opt转换为 paddle lite 格式, 但是在转换过程中出现了以下错误

opt_linux2.7 --model_dir=PyramidBox_WiderFace
[I  4/ 1 11:24: 1.198 /island/Paddle-Lite/lite/api/cxx_api.cc Build:262] Load model from file.
[I  4/ 1 11:24: 2.461 ...island/Paddle-Lite/lite/core/optimizer.h:259 RunPasses] == Running pass: lite_quant_dequant_fuse_pass
[I  4/ 1 11:24: 2.490 ...island/Paddle-Lite/lite/core/optimizer.h:283 RunPasses] == Finished running: lite_quant_dequant_fuse_pass
[I  4/ 1 11:24: 2.490 ...island/Paddle-Lite/lite/core/optimizer.h:259 RunPasses] == Running pass: weight_quantization_preprocess_pass
[I  4/ 1 11:24: 2.491 ...island/Paddle-Lite/lite/core/optimizer.h:283 RunPasses] == Finished running: weight_quantization_preprocess_pass
[I  4/ 1 11:24: 2.491 ...island/Paddle-Lite/lite/core/optimizer.h:259 RunPasses] == Running pass: lite_conv_elementwise_fuse_pass
[I  4/ 1 11:24: 2.514 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 67 subgraph
[I  4/ 1 11:24: 2.518 ...island/Paddle-Lite/lite/core/optimizer.h:283 RunPasses] == Finished running: lite_conv_elementwise_fuse_pass
[I  4/ 1 11:24: 2.518 ...island/Paddle-Lite/lite/core/optimizer.h:259 RunPasses] == Running pass: lite_conv_bn_fuse_pass
[I  4/ 1 11:24: 2.539 ...le-Lite/lite/core/mir/pattern_matcher.cc:108 operator()] detected 24 subgraph
[I  4/ 1 11:24: 2.546 ...island/Paddle-Lite/lite/core/optimizer.h:283 RunPasses] == Finished running: lite_conv_bn_fuse_pass
[I  4/ 1 11:24: 2.546 ...island/Paddle-Lite/lite/core/optimizer.h:259 RunPasses] == Running pass: lite_conv_elementwise_fuse_pass
[I  4/ 1 11:24: 2.564 ...island/Paddle-Lite/lite/core/optimizer.h:283 RunPasses] == Finished running: lite_conv_elementwise_fuse_pass
[I  4/ 1 11:24: 2.564 ...island/Paddle-Lite/lite/core/optimizer.h:259 RunPasses] == Running pass: lite_conv_conv_fuse_pass
[F  4/ 1 11:24: 2.565 ...island/Paddle-Lite/lite/core/mir/node.cc:32 picked_kernel] Check failed: !valid_kernels_.empty(): no kernel for depthwise_conv2d_transpose
Aborted

Opt 的 2.7 和 2.8 版本我都试过, 全都报这个错误 Check failed: !valid_kernels_.empty(): no kernel for depthwise_conv2d_transpose

请问有什么解决方法吗?