Training SVM

[MarcoSaku]

Hi everybody, I am implementing the paper about the deep-poselets, it is the next paper published by Bourdev and Malik (http://arxiv.org/pdf/1407.0717v1.pdf). I created the deep-net and I trained it as mentioned in the paper. Now I have to use it in training part to extract feature-vector instead the HOG. Could I have the code about the training part? It could be useful for me also the 2000 poselets candidates (obtained by H3D). In this way replacing the 256 d-array instead the array given by HOG i can complete my implementation. Thanks for the attention, Marco Saviano

[gkioxari]

Hi Marco,

Since you are referring to the deep-poselet work and are interested in reimplementing that one, you should probably contact Lubomir. K-poselets is our implementation of poselets and might be different than Lubomir's implementation.

For k-poselets, you can find code for the SVM training in code/training/train_kposelet.m. We also provide our own list of poselets in http://www.cs.berkeley.edu/~gkioxari/kposelets/. For the 2000 poselets of the deep-poselets, you should contact Lubomir.

[zhangjingling]

when I run the demo_bbox_dection (),i get the error : Undefined function 'collect_detections_on_imglist' for input arguments of type 'struct'. Error in demo_bbox_detection (line 16)

Can you help me? I use matlab 2013a,win7