This folder contains codes, data, and models relating to an implementation of: -- Dinesh Jayaraman and Kristen Grauman, "Look-ahead before you leap: end-to-end active recognition by forecasting the effect of motion", ECCV 2016. (http://vision.cs.utexas.edu/projects/lookahead_active/lookahead_active_cvpr16.pdf) To evalute a pre-trained model e.g. models/2002782.dat, run: $ th evaluate_model.lua --jobno 2002782 --evalFlag --rho 3 --num_repeats 10 (see various options and meanings in source) To train a new model, run: $ th lookahead_active.lua (see various options and meanings in source) Enclosed models names: models/2002782.dat - 3-step model models/2002450.dat - 1-step model (used to initialize during training of 3-step model, with the initModel option of lookahead_active.lua) At this point, the contents of this folder have been specialized for the GERMS dataset. Torch packages you may have to install: --------------------------------------- ElementResearch/rnn - commit cc1e1c2 nicholas-leonard/dp (if trouble, try commits close to rnn's cc1e1c2 commit in time) ElementResearch/dpnn (if trouble, try commits close to rnn's cc1e1c2 commit in time) torch/optim keplerproject/luafilesystem e-lab/torch-toolbox (for Weight-init.lua) torch/image deepmind/torch-hdf5 torch/gnuplot