search

facebookarchive / C3D

C3D is a modified version of BVLC caffe to support 3D ConvNets.
Other
1.16k stars 506 forks source link

the conv3d_ucf101_train.prototxt file is not same as you described in the paper #94

Closed learntolearn11 closed 8 years ago

learntolearn11 commented 8 years ago

Hi, I see the conv3d_ucf101_train.prototxt, there are only five layers named conv1a,conv2a,conv3a,conv4a,conv5a,while in your paper 'Learning Spatiotemporal Features with 3D Convolutional Networks', there are 8 layers,named conv1a,conv2a,conv2b,conv3a,conv3b,conv4a,conv4b,conv5a,conv5b. so which is better?

dutran commented 8 years ago

This is the experiments for smaller networks showed in Fig 2. Not the one in Fig 3.

learntolearn11 commented 8 years ago

which is better?

dutran commented 8 years ago

@learntolearn11 Please read the paper for full details. Thanks.