Table 1.
Three sub-tables show the top-1, top-2, and top-3 accuracies of categorizing individual movie frames by using decoders trained with data from the same (intra-subject) or different (inter-subject) subject. Each row shows the categorization accuracy with the decoder trained with a specific subject's training data; each column shows the categorization accuracy with a specific subject's testing data and different subjects’ decoders. The accuracy was quantified as the percentage by which individual movie frames were successfully categorized as one of the top-1, top-2, or top-3 categories. The accuracy was also quantified as a fraction number (shown next to the percentage number): the number of correctly categorized frames over the total number of frames that could be labeled by the 15 categories (N = 214 for 1 8-min testing movie)
| Decoding accuracy for the semantic descriptions of a novel movie | ||||
|---|---|---|---|---|
| Train/test | Subject 1 | Subject 2 | Subject 3 | |
| Top-1 | subject 1 | 42.52% (91/214) | 24.30% (52/214) | 23.83% (51/214) |
| subject 2 | 20.09% (43/214) | 50.47% (108/214) | 22.90% (49/214) | |
| subject 3 | 24.77% (53/214) | 33.64% (72/214) | 50.00% (107/214) | |
| Top-2 | subject 1 | 59.81% (128/214) | 41.12% (88/214) | 43.93% (94/214) |
| subject 2 | 35.51% (76/214) | 70.09% (150/214) | 35.98% (77/214) | |
| subject 3 | 41.12% (88/214) | 42.06% (90/214) | 66.36% (142/214) | |
| Top-3 | subject 1 | 67.76% (145/214) | 55.14% (118/214) | 53.27% (114/214) |
| subject 2 | 48.13% (103/214) | 74.77% (160/214) | 50.93% (109/214) | |
| subject 3 | 50.93% (109/214) | 52.34% (112/214) | 72.90% (156/214) | |