From 195ce76b842048cd8a5809b0fe362a101b1de556 Mon Sep 17 00:00:00 2001
From: Manuel Gunther <siebenkopf@googlemail.com>
Date: Mon, 23 Oct 2017 18:15:56 -0600
Subject: [PATCH] Implemented recursive scanning for RR and DIR (not tested
yet)
---
bob/bio/base/script/collect_results.py | 86 +++++++++++++++++---------
1 file changed, 57 insertions(+), 29 deletions(-)
diff --git a/bob/bio/base/script/collect_results.py b/bob/bio/base/script/collect_results.py
index 1aacf257..b4d9ed04 100644
--- a/bob/bio/base/script/collect_results.py
+++ b/bob/bio/base/script/collect_results.py
@@ -27,6 +27,7 @@ The measure type of the development set can be changed to compute "HTER" or
import sys, os, glob
import argparse
+import numpy
import bob.measure
import bob.core
@@ -47,11 +48,10 @@ def command_line_arguments(command_line_parameters):
parser.add_argument('-s', '--sort', action='store_true', help = "Sort the results")
parser.add_argument('-k', '--sort-key', dest='key', default = 'nonorm-dev', choices= ('nonorm-dev','nonorm-eval','ztnorm-dev','ztnorm-eval','dir'),
help = "Sort the results according to the given key")
- parser.add_argument('-c', '--criterion', dest='criterion', default = 'EER', choices = ('EER', 'HTER', 'FAR'),
+ parser.add_argument('-c', '--criterion', dest='criterion', default = 'EER', choices = ('EER', 'HTER', 'FAR', 'RR', 'DIR'),
help = "Minimize the threshold on the development set according to the given criterion")
parser.add_argument('-o', '--output', help = "Name of the output file that will contain the EER/HTER scores")
- parser.add_argument('--parser', default = '4column', choices = ('4column', '5column'), help="The style of the resulting score files; rarely changed")
parser.add_argument('--self-test', action='store_true', help=argparse.SUPPRESS)
@@ -62,9 +62,6 @@ def command_line_arguments(command_line_parameters):
bob.core.log.set_verbosity_level(logger, args.verbose)
- # assign the score file parser
- args.parser = {'4column' : bob.measure.load.split_four_column, '5column' : bob.measure.load.split_five_column}[args.parser]
-
return args
class Result:
@@ -79,32 +76,63 @@ class Result:
def _calculate(self, dev_file, eval_file = None):
"""Calculates the EER and HTER or FRR based on the threshold criterion."""
- dev_neg, dev_pos = self.m_args.parser(dev_file)
-
- # switch which threshold function to use;
- # THIS f***ing piece of code really is what python authors propose:
- threshold = {
- 'EER' : bob.measure.eer_threshold,
- 'HTER' : bob.measure.min_hter_threshold,
- 'FAR' : bob.measure.far_threshold
- } [self.m_args.criterion](dev_neg, dev_pos)
-
- # compute far and frr for the given threshold
- dev_far, dev_frr = bob.measure.farfrr(dev_neg, dev_pos, threshold)
- dev_hter = (dev_far + dev_frr)/2.0
-
- if eval_file:
- eval_neg, eval_pos = self.m_args.parser(eval_file)
- eval_far, eval_frr = bob.measure.farfrr(eval_neg, eval_pos, threshold)
- eval_hter = (eval_far + eval_frr)/2.0
- else:
- eval_hter = None
- eval_frr = None
+ if self.m_args.criterion in ("RR", "DIR"):
+ scores_dev = bob.measure.load.cmc(dev_file)
+ if eval_file is not None:
+ scores_eval = bob.measure.load.cmc(eval_file)
+
+ if self.m_args.criterion == "DIR":
+ # get negatives without positives
+ negatives = [max(neg) for neg, pos in scores_dev if (pos is None or not numpy.array(pos).size) and neg is not None]
+ if not negatives:
+ raise ValueError("There need to be at least one pair with only negative scores")
+ threshold = bob.measure.far_threshold(negatives, [], 0.001)
+ DIR_dev = bob.measure.detection_identification_rate(scores_dev, threshold)
+ if eval_file is not None:
+ # re-compute the threshold for eval file
+ negatives = [max(neg) for neg, pos in scores_eval if (pos is None or not numpy.array(pos).size) and neg is not None]
+ if not negatives:
+ raise ValueError("There need to be at least one pair with only negative scores")
+ threshold = bob.measure.far_threshold(negatives, [], 0.001)
+ DIR_dev = bob.measure.detection_identification_rate(scores_eval, threshold)
+ else:
+ DIR_eval = None
+ return DIR_dev, DIR_eval
+
+ else:
+ # Recognition Rate
+ RR_dev = bob.measure.recognition_rate(scores_dev)
+ RR_eval = None if eval_file is None else bob.measure.recognition_rate(scores_eval)
+ return RR_dev, RR_eval
- if self.m_args.criterion == 'FAR':
- return (dev_frr, eval_frr)
else:
- return (dev_hter, eval_hter)
+
+ dev_neg, dev_pos = bob.measure.load.split(dev_file)
+
+ # switch which threshold function to use;
+ # THIS f***ing piece of code really is what python authors propose:
+ threshold = {
+ 'EER' : bob.measure.eer_threshold,
+ 'HTER' : bob.measure.min_hter_threshold,
+ 'FAR' : bob.measure.far_threshold
+ } [self.m_args.criterion](dev_neg, dev_pos)
+
+ # compute far and frr for the given threshold
+ dev_far, dev_frr = bob.measure.farfrr(dev_neg, dev_pos, threshold)
+ dev_hter = (dev_far + dev_frr)/2.0
+
+ if eval_file:
+ eval_neg, eval_pos = bob.measure.load.split(eval_file)
+ eval_far, eval_frr = bob.measure.farfrr(eval_neg, eval_pos, threshold)
+ eval_hter = (eval_far + eval_frr)/2.0
+ else:
+ eval_hter = None
+ eval_frr = None
+
+ if self.m_args.criterion == 'FAR':
+ return (dev_frr, eval_frr)
+ else:
+ return (dev_hter, eval_hter)
def nonorm(self, dev_file, eval_file = None):
self.nonorm_dev, self.nonorm_eval = self._calculate(dev_file, eval_file)
--
GitLab