-
…. The C API and command line tool are updated to support the new function. Changes are: 1. Change how operating points are computed to mirror eval (i.e. operating points are added only if both the true positive AND false positive count changes). This fixes a bug where successive operating points could have the same FAR, which caused nan results in the output. 2. Change the optional pdf argument to a CSV argument (again mirroring eval) so that results are written as CSVs and can be combined to show multiple results on the same plot
-
This reverts commit 694a3826d0d03713b2070c40431f8ecf3d0b3663.
-
Updates to eval_detection
-
…errors out if result of Evaluate() does not meet or exceed metric.
-
Surely this will save many bytes of source code.
-
…rom ground truth bounding boxes.
-
Only verification mode evaluation is supported. The main issue is to avoid loading potentially very large similarity and mask matrices into main memory.
-
Defaulting to ClusterID did not make much sense, since the key being indexed is in fact the ground truth, not the cluster labels (cluster labels are stored as indices in a csv file, not a gallery). Also, use Format to read similarity matrices instead of assuming mtx is in use.
-
Instead of returning references to static memory (which are invalidated by subsequent or concurrent calls to the function), require users to input string buffers to functions that return strings. Affected functions: br_get_filename br_get_metadata_string br_scratch_path br_objects br_most_recent_message In some GUI functions, use C++ functions instead of going through the c api add a c++ method correspondign to br_objects (which returns a stringlist instead of merging the lines) br_read_pipe and br_get_header also use static memory in a similar way, but are not addressed in this commit.