Table 3 Confusion Matrices at participant level: for EgoTerrainNet-Outdoor and -Indoor, MobileNetV2’s pre-trained on ImageNet dataset were fine-tuned. The validation accuracies (during training) for -Outdoor and -Indoor versions were 99.23 and 85.26, respectively. \(\star :\) camera was unintentionally mounted upside-down by the participants or was set to take photos (not videos), \(\dagger :\) Participants living in the same home, HFM: high-friction materials, \(\bigtriangleup\): cases that are discussed in "Deeper analysis of lower accuracies ".Darker shades of gray indicate higher per-class accuracies
