Skip to main content
. 2024 Apr 29;17(3):10.16910/jemr.17.3.3. doi: 10.16910/jemr.17.3.3

Figure 4.

Figure 4.

Samples of contentious frames: (a) & (b) Both models tended to mislabel the marker on the tablet’s edge, where manual labelling is also difficult. (c) & (d) Model v2 mislabeled other humans holding the tablet. (e) Concrete-only backgrounds were mislabeled more often than vegetation-backgrounds. (f) A red car in the back, no model detected the (also red) gaze marker.