Some code packages implementing my recent papers. Questions? Contact me jhoffman _at_ eecs.berkeley.edu.
|LSDA||Large scale detection through adaptation. A model that adapts classification models into detectors. Release of a >7.5K category detector.||Caffe|
|Caffe||Caffe is a sofeware package for training, finetuning, and evaluating CNNs for classification and detection. A faster and more flexible version of DeCAF.||-|
|DeCAF||DeCAF is a deep learning based feature (Donahue ICML`14). Check out the live demo here.||-|
|LS-MMDT||Faster implementation of MMDT which is integrated with a liblinear base package. This version should be used for large scale data sets. Makes use of the optimization algorithm presented at ICCV and NIPS recently (Rodner ICCV`13).||-|
|MMDT||Learns a category invariant transformation using max-margin constraints. Faster and better performance than da-transforms (Hoffman ICLR`13).||liblinear-weights, arc-t|
|Domain Discovery||Uses category constraints to automatically separate a heterogenous dataset into cohesive domains (Hoffman ECCV`12).||libsvm, arc-t|
|ARC-t||Learns a category invariant transform using similarity constraints. (Saenko ECCV`10, Kulis CVPR`11)||-|