image_recognition_analysis

Tool(s) for image recognition

Classifier_metrics

Usage:

rosrun image_recognition_analysis classifier_metrics results.csv

The classifier_metrics script produces a plot that gives a good view of the performance of a given classifier. As input, it takes a .csv file of the following format:

• First row are column headers
• Other rows are 1 per image
• There is a ground_truth column
• For each output label the classifier, there must be a column with the weight of the corresponding class

e.g.

ground_truth,cat,dog
cat,0.9,0.1
dog,0.2,0.8
dog,0.3,0.7

To use this on e.g. a tensorflow model:

cd path/to/dir/with/evaluation_set # directory with a subdir for each class, as with training
rosrun image_recognition_tensorflow evaluate_classifier output_graph.pb $output_labels.txt -o result.csv .
rosrun image_recognition_analysis classifier_metrics result.csv

Changelog for package image_recognition_analysis

0.0.5 (2019-06-08)

• Added catkin lint to all packages
• refactor package xml to 2.0
• rename ROS pkgs with image_recognition_prefix
• fix(confusion_matrix): Use correct labels on axis
• Refactor _tools to _analysis and restructured code
• Contributors: Loy, Rein Appeldoorn