Generated output
The software will create a new sub-folder within the selected results
folder for each input video. Inside each sub-folder the following
outputs are stored by the software:
- all_events_unfiltered: Inside this csv file, all detected
events are listed containing the video timestamp, the bee ID, the
event type (entering or leaving) and the cavity ID. This list is
completely unfiltered and may contain errors.
- error_corrected_events: This csv file contains all events
that remain after error correction: the software identifies missing
events within sequences of enter – leave – enter. Sequences with
missing events are not considered for the creation of below described
output files (address_book, nest_recognition, flight_list). The
file contains the video timestamp, the bee ID, the cavity ID and the
type of event (entering or leaving a cavity). Additionally, it is
indicated for each event whether it was used for the output files
address_book, nest_recognition and flight_list. Note that some
events might be missing in these files due to the strict error
correction of the software.
- address_book: This csv file contains all bees that were
assigned to a nest and lists the according bee and cavity IDs. This
data (assignments between individual bees and the cavity (or cavities,
respectively) they are nesting in) is of interest for assessments of
nesting progress and reproductive success of individual nesting
females. In order to assign only cavities to females which are used
for nesting (in contrast to simply probed cavities not used for
nesting), a cavity is only assigned to an individual bee if (i) the
bee stays inside the cavity for a time span that is minimally required
by a nesting bee to unload collected pollen for offspring provision,
and (ii) the bee does not enter another cavity during a time span that
is minimally required by a bee to collect pollen or material such as
mud for nest construction (e.g., construction of brood-cell walls).
The default setting of these two time spans are both 40 s in the
published open-source version of the software. These values were
chosen based on over 20 h of direct observation of Osmia
bicornis females nesting in a natural habitat in Switzerland (Bättig
D., unpublished data). However, the species under study or
experimental setting may require adjustment of these threshold values.
This can be done in the “config” file of the software, which can be
selected as an optional input file for the analysis (see software
manual in the Supporting Information).
Nesting progress, i.e. the number of produced brood cells and offspring,
can be tracked by repeatedly photographing the nest cavities (Fig. 1),
e.g. before and after an assessment day. Linking this data with theaddress_book file (created form a video recorded on the same
assessment day) based on cavity IDs permits to measure individual per
female reproductive success for this time period.
nest_recognition: This csv file contains the number of
cavities a female enters before finding its nest (i.e. number of
probed “wrong” cavities before finding the “correct” nesting
cavity). Besides the bee ID and the number of probed cavities, the
file also lists the video timestamp.
flight_list: This csv file provides flight durations of
individual females from leaving the nesting cavity until returning to
it again (i.e., foraging trip or mud collection duration). Besides the
bee ID and the flight duration the file also lists the video
timestamp.
If of interest, flight activity can be assessed by classifying females
that perform flights as active and are therefore listed in the
flight_list file. For this measurement, the number of total, alive
females needs to be known however, which can be assessed by taking
pictures of the nest layer (Fig. 1) during the night when females are
roosting inside cavities.
6. visualization: Through the “visualize results” option a
video file in mp4 format can be created with all detected events
visualized. This file can be used to visually check the performance of
the software and to find potential errors, which can be used to retrain
the software (see below) and improve the precision.