search

jiangtaoxie / MPN-COV

@ICCV2017: For exploiting second-order statistics, we propose Matrix Power Normalized Covariance pooling (MPN-COV) ConvNets, different from and outperforming those using global average pooling.
http://peihuali.org/iSQRT-COV/index.html
MIT License
75 stars 27 forks source link

error raised as following when fine-tuning is completed #2

Closed xinxliu closed 6 years ago

xinxliu commented 7 years ago 
val: epoch 100:  48/ 49: 22.4 (22.4) Hz objective: 0.519 top1error: 0.127 top5error: 0.028
val: epoch 100:  49/ 49: 22.4 (20.3) Hz objective: 0.521 top1error: 0.128 top5error: 0.028
Subscripted assignment between dissimilar structures.

Error in dagnn.DagNN/addParam (line 414)
      obj.params(p) = struct(...

Error in dagnn.Layer/attach (line 236)
        net.addParam(char(param)) ;

Error in dagnn.DagNN/addLayer (line 34)
obj.layers(index).block.attach(obj, index) ;

Error in cnn_imagenet_deploy>dagMergeBatchNorm (line 127)
    net.addLayer(playerName, ...

Error in cnn_imagenet_deploy (line 20)
  dagMergeBatchNorm(net) ;

Error in MPN_COV_main_ft (line 123)
net = cnn_imagenet_deploy(net) ;

Error in run (line 96)
evalin('caller', [script ';']);

So will this error influence the model and why it occured? Thank u!

jiangtaoxie commented 7 years ago

will this error influence the model No. why it occured? When the training is finished, the function 'cnn_imagenet_deploy.m' will convert the original model(E.g 'net-epoch-100.mat') to 'deployed model' . It's not a important issue, but I'll check and fix it

xinxliu commented 7 years ago

Thank you for the reply.