Open y22ma opened 8 years ago
Found the paper, which indicates that the network has two 7x7x32 layers with a L2-Reg layer in between and random connection. And the network is trained with CIFAR datasets with K-means unsupervised learning.
Will try to replicate a model and training scripts to see if I can reproduce the results, and create a PR once I get them working.
Just wondering where I can find information regarding the net design/architecture for the NN used in the 'tracker' demo?