Tests are run using GSE job manager:
on DAS4: qsub -v CFGFILE=foo -v TMPFOLDER=bar testjob.job
on own computer : ./testjob.sh $CFGFILE $TMPFOLDER
The code expects a python package called nbnn (Found here )
voc_cls.sh and voc_cls.job Main scripts for running tests for classification on full images (VOC)
voc_cls.py and voc_cls_test.py Main files that respectively train and test on full images (VOC)
voc_cls_bbox.sh and voc_cls_bbox.job Main scripts for running tests for classification on bboxes of images (VOC)
voc_cls_bbox.py and voc_cls_bbox_test.py Main files that respectively train and test on bboxes of images (VOC)
utils.py Helper functions
procedures.py Generic main routines for training (and testing)
setlog.sh
shell script to make a temporary log config file, used by init\_log() in utils.py
osx_cls.cfg, osx_cls_bbox.cfg, das_cls.cfg and das_cls_bbox.cfg config files for tests on laptop (osx) and server (das)
parse_log.py
usage: $ python parse_log.py logfile.log
Gives the lines in the log where most memory is used
cal_perf.py
usage: $ python cal_perf.py confmat.txt
Calculates the MRR and plots the confmat of the results of a caltech101 test
blank.log.cfg
Generic log config file that is used (copied) by setlog.cfg
README.markdown This file
Deprecated: need to write new ones
detection_cheap.py
Usage: $ python detection_cheap.py some/config/file.cfg
Main file for a cheap kind of detection on the VOC11 data set.
Shell script that searches for a file called pattern_1.cfg in path some/config
and duplicates this file N times. During duplication, _1 patterns within the file
are replaced with _i. Useful in case multiple iterations with the same settings
need to be run.
pascal.py Dataset classes for the Pascal Detection (future: Classification) tasks, that extend nbnn's VOC Dataset class. Also contains a VOCResultsHandler class that interprets and saves VOC detection results