Incubator-mxnet: Finetune error according tutorial: ValueError: Cannot find output that matches name "flatten0_output"

[ysh329]

I want to finetune pretrained inception-BN, squeezeNet, NIN, caffenet, vgg16, vgg19 models from model zoo, which happened this ValueError.

It's okay for finetuning resnet, resneXt models.

@kevinthesun @zackchase @sunilmallya @aileli :laughing:
THanks a lot!

[sunbinbin1991]

I finetune a model have same question with you

[ysh329]

@sunbinbin1991 Do you add flatten0_output layer, you can have a try.

[ysh329]

@tornadomeet :sos:

[sunbinbin1991]

I find my problem, because I used caffenet to finetune my model, actually, this model have no layer"flatten0",so instead, change it with "drop7", that works, in a demo ,it said:

layer_name: the layer name before the last fully-connected layer

It depends, after change it ,it works ,good luck

[ysh329]

@sunbinbin1991 okay, my brother, Thanks.

[tornadomeet]

@ysh329 have you solved it ?

[ysh329]

@tornadomeet Yes, Thanks a lot!

After attaining the sym varible using load_checkpoint, I used sym.get_internals().list_outputs() to check the outputs of each layer. And replace flatten0 with flatten_0.