toolkit1.3.0+UbuntuPC模拟器+Pytorch DenseNet出错

环境：toolkit1.3.0+UbuntuPC模拟器+Pytorch转换。
问题描述：在examples/pytorch/test.py中，我们将resnet18替换成densenet201，发现不能运行成功。rknn.init_runtime()运行过程中显示Restarting kernel...

jefferyzhang 发表于 2020-3-17 15:04
pytorch版本？
log？

python3.5，pytorch 1.2.0+cpu。
rknn.init_runtime出问题的log如下：
I [setup_nodes:4017]name=convolution_at_9310_36
I [setup_nodes:4017]name=cat_at_input579.1_331
I [setup_nodes:4017]name=batch_norm_at_input580.1_250
I [setup_nodes:4017]name=relu__at_input581.1_169
I [setup_nodes:4017]name=convolution_at_input582.1_136
I [setup_nodes:4017]name=relu__at_input584.1_70
I [setup_nodes:4017]name=convolution_at_9388_37
I [setup_nodes:4017]name=cat_at_input585.1_332
I [setup_nodes:4017]name=batch_norm_at_input586.1_251
I [setup_nodes:4017]name=relu__at_input587.1_170
I [setup_nodes:4017]name=convolution_at_input588.1_137
I [setup_nodes:4017]name=relu__at_input590.1_71
I [setup_nodes:4017]name=convolution_at_9467_38
I [setup_nodes:4017]name=cat_at_input591.1_333
I [setup_nodes:4017]name=batch_norm_at_input592.1_252
I [setup_nodes:4017]name=relu__at_input593.1_171
I [setup_nodes:4017]name=convolution_at_input594.1_138
I [setup_nodes:4017]name=relu__at_input596.1_72
I [setup_nodes:4017]name=convolution_at_9547_39
I [setup_nodes:4017]name=cat_at_input597.1_334
I [setup_nodes:4017]name=batch_norm_at_input598.1_253
I [setup_nodes:4017]name=relu__at_input599.1_172
I [setup_nodes:4017]name=convolution_at_input600.1_139
I [setup_nodes:4017]name=relu__at_input602.1_73
I [setup_nodes:4017]name=convolution_at_new_features2.1_40
I [setup_nodes:4017]name=cat_at_input603.1_7
I [setup_nodes:4017]name=batch_norm_at_input604.1_6
I [setup_nodes:4017]name=relu__at_input605.1_5
I [setup_nodes:4017]name=adaptive_avg_pool2d_at_out.1_4
I [setup_nodes:4017]name=permute_3
I [setup_nodes:4017]name=trans_flatten_at_input606.1_2
I [setup_nodes:4017]name=addmm_at_9685_1
I [vnn_CreateRKNN:5249]Norm tensor initialize
I [vnn_CreateRKNN:5264]Const tensor initialize
I [vnn_CreateRKNN:5279]Virtual tensor initialize

Restarting kernel...
之前的步骤ret返回0，以上log还没保存至verbose_file，就把python kernel整挂了

jefferyzhang 发表于 2020-3-17 15:04
pytorch版本？
log？

另，我们采用的torchvision中的原生densenet201模型，
import torchvision.models as models
net = models.densenet201(pretrained=True)
谢谢

jefferyzhang 发表于 2020-3-17 15:20
问题已提交

期待你们的反馈！！！