jcas
Matlab C++
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
@jcas
Analysis
BCFWstruct
colormaps
toolboxes
.gitignore
Compute_Statistics.m
Compute_Statistics_with_Bootstrapping.m
Initialization.m
InitializationWCVPPC.m
Initialization_Bertrand.m
Mac.m
README.txt
Readme.md
alpha_expansion_intersection.m
alpha_expansion_labelcost.m
build_aggregated_superpixels_histograms.m
build_dictionary_unary.m
build_superpixels_histograms.m
build_superpixels_histograms_in_parallel.m
build_topdown_dictionary.m
build_trainingset_unary_histogramsvm.m
compute_IP_adj.m
compute_intersection_kernel.m
compute_label_histograms.m
compute_pairwise_cost.m
compute_topdown_unaries.m
compute_unary_costs.m
constraintFnCP.m
cutting_plane_learning.m
delete_files.m
display_error.m
extract_features.m
extract_features_in_parallel.m
featureFnCP.m
generate_topdown_descriptors.m
get_class_specific_textonboost_unary.m
get_dataset_path.m
get_ground_truth.m
infer_words.m
latent_duality_gap.m
latent_solverBCFWpos.m
latent_svm_struct_mod.m
latent_svm_struct_mod_1slack.m
load_and_analyze_results.m
load_precomputed_unary.m
lossFnCP.m
mySSVM.m
my_latent_SSVM.m
read_file.m
run_experiments.m
run_solver.m
save_TBunary_files.m
setup_data.m
setup_data_wcvp.m
sp_label_to_pixlabel.m
svm_struct_mod.m
svm_struct_mod_1slack.m
test_kernel_svm.m
train_in_parallel.m
train_kernel_svm.m
train_unary_classifiers.m
visualize_results.m
wordsFnCP.m

Readme.md

Readme JCaS

I. Installation In order to use the code for Joint Categorization and segmentation, you have to install the following toolboxes: _ blocks (0.1.1 or above) _ vlfeat (0.9.18 or above) _ maxflow-v3.01 or above (Miki Rubinstein Wrapper available on Matlab FileExchange) _ graphAnalysisToolbox-1.0 or above _ GCMex wrapper for maxflow. NB : Don't forget to add them in Matlab's search path

II. Usage

(0) Everything is parametrized in a JCaS() object. You can create one and explore the structure and the options available.

a.Image/ground truth format : _ Put the images in a directory _ Put the ground truth labelings in another one, with .mat format being a matlab array of integers from 1 to the number of classes, and if it exists 0 as the void class.

b. Adding the database: _ Add your database ine the @jcas/makedb.m file and

c. Parameters _ All parameters can be modified in the Initialization.m file. To run the code without any further modification, just run the script. _ You can transparently change the parameters, and the code will take care of reusing what was previously computed to save computation time/storage space.

d. Force recomputation If you want to recompute some part of the algorithm,

e. Further modification Most of the options are in a single file. You can add your own superpixels/unary/topdown features in the compute*.m files in @jcas dir.

III. Modes You can change the jcas.mode option to the following : 0 = Unary only 1 = Unary and pairwise 2 = Unary + pairwise + linear topdown from ECCV12 paper 3 = Unary + pairwise + linear topdown from ECCV + label cost 4 = Unary + pairwise + linear topdown + label cost (= topdown histogram norm) 5 = Unary + pairwise + label cost only 6 = Unary + pairwise + intersection kernel (PAMI) 7 = unary + pairwise + linear topdown + Unary on words (ECCV) 8 = unary + pairwise + linear topdown + CRF on words (ECCV) (Under construction)