Encounter error when trying to use ResNet-50 to replace VGG-19 as the loss network
michaelhuang74 opened this issue · comments
I tried to replace the VGG-19 with ReSNet-50 for the loss network used in the training.
I downloaded the prototxt and caffemodel files from https://onedrive.live.com/?authkey=%21AAFW2-FVoxeVRck&id=4006CBB8476FF777%2117887&cid=4006CBB8476FF777, which is given on https://github.com/KaimingHe/deep-residual-networks
Then I tried to launch the training as following:
th train.lua -style_image style/face_bottle_o.png -style_size 512 -image_size 256 -style_layers res2b_relu,res3b_relu,res4b_relu,res5b_relu -content_layers res4b_relu -style_weight 20 -content_weight 1 -proto_file data/pretrained/ResNet-50-deploy.prototxt -model_file data/pretrained/ResNet-50-model.caffemodel -checkpoints_path checkpoint/ -checkpoints_name face_bottle_o.s512i256.sL2b3b4b5b.cL4b.sw20.cw1.ResNet50.4x -num_iterations 10000 -batch_size 4 -save_every 2000
I encountered the following errors.
torch.display not found. unable to plot
Using TV loss with weight 1e-06
[libprotobuf ERROR google/protobuf/text_format.cc:245] Error parsing text-format caffe.NetParameter: 26:26: Message type "caffe.LayerParameter" has no field named "batch_norm_param".
Successfully loaded data/pretrained/ResNet-50-model.caffemodel
warning: module 'bn_conv1 [type BatchNorm]' not found
warning: module 'scale_conv1 [type Scale]' not found
warning: module 'pool1_pool1_0_split [type Split]' not found
warning: module 'bn2a_branch1 [type BatchNorm]' not found
warning: module 'scale2a_branch1 [type Scale]' not found
warning: module 'bn2a_branch2a [type BatchNorm]' not found
warning: module 'scale2a_branch2a [type Scale]' not found
..........(similar warning messages)
warning: module 'bn5c_branch2b [type BatchNorm]' not found
warning: module 'scale5c_branch2b [type Scale]' not found
warning: module 'bn5c_branch2c [type BatchNorm]' not found
warning: module 'scale5c_branch2c [type Scale]' not found
warning: module 'res5c [type Eltwise]' not found
conv1: 64 3 7 7
res2a_branch1: 256 64 1 1
Segmentation fault (core dumped)
Any idea how to resolve this issue? Thanks.
Hello, I've never tried to use resnet and have no clue for this segfault...
Hello @michaelhuang74 . There was title changes in layers and it's attributes names.
I have faced similar problem soon.
Shortly: try changing in prototxt 'batch_norm_param' to 'bn_param'
There can be more than 1 name error