Instruction to run DPM v5 with Pascal VOC devkit Assume you will to put everything inside a path called rootVOC4SUN (e.g. /home/jxiao/expVOC4SUN) 1. Download DPM v5 (http://people.cs.uchicago.edu/~rbg/latent/voc-release5.tgz) from http://people.cs.uchicago.edu/~rbg/latent/ 2. Untar voc-release5.tgz to rootVOC4SUN/voc-release5/ (i.e. you should have the file rootVOC4SUN/voc-release5/compile.m by now) In matlab, cd into rootVOC4SUN/voc-release5/, and run >> addpath(genpath('.')); compile as instructed by rootVOC4SUN/voc-release5/README 3. Download Pascal VOC devkit 2012 (http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2012/VOCdevkit_18-May-2011.tar) from http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2012/index.html 4. Untar VOCdevkit_18-May-2011.tar to rootVOC4SUN/VOC2012/VOCdevkit (i.e. you should have the folder rootVOC4SUN/VOC2012/VOCdevkit/VOCcode by now) 5. Edit VOCinit.m under rootVOC4SUN/VOC2012/VOCdevkit/VOCcode Line 9: From: VOCopts.dataset='VOC2012'; To: VOCopts.dataset='SUN2012pascalformat'; Line 30-31: From: VOCopts.testset='val'; % use validation data for development test set % VOCopts.testset='test'; % use test set for final challenge To: % VOCopts.testset='val'; % use validation data for development test set VOCopts.testset='test'; % use test set for final challenge 6. Edit VOCreadrecxml.m under rootVOC4SUN/VOC2012/VOCdevkit/VOCcode Line 9-10: From: rec.source.annotation=x.source.annotation; rec.source.image=x.source.image; To: %rec.source.annotation=x.source.annotation; %rec.source.image=x.source.image; 7. Edit voc_config.m under rootVOC4SUN/voc-release5/ Line 22: From: BASE_DIR = '/var/tmp/rbg'; To: BASE_DIR = 'rootVOC4SUN'; % your rootVOC4SUN. Without a "/" at the end. Line 25: From: PASCAL_YEAR = '2007'; To: PASCAL_YEAR = '2012'; Line 149: From: conf = cv(conf, 'training.train_set_fg', 'trainval'); To: conf = cv(conf, 'training.train_set_fg', 'train'); 8. Download SUN Database 2012 release in Pascal Format ( http://groups.csail.mit.edu/vision/SUN/releases/SUN2012pascalformat.tar.gz ) from http://sundatabase.mit.edu/ 9. Untar SUN2012pascalformat.tar.gz to rootVOC4SUN/VOC2012/VOCdevkit/SUN2012pascalformat (i.e. you should have the file rootVOC4SUN/VOC2012/VOCdevkit/SUN2012pascalformat/report.txt by now) 10. Now, everything is ready, and you can train a detector by: In matlab, cd into rootVOC4SUN/voc-release5/ >> pascal('mug',3); % this will train a mug detector using SUN2012. You can see the list of object category in rootVOC4SUN/VOC2012/VOCdevkit/SUN2012pascalformat/report.txt . ------------------------- Notice: You can download our pretrained result at http://sundatabase.mit.edu/ We welcome any feedback or bug report. Please email me (http://mit.edu/jxiao/) Jianxiong Xiao