Table 2.

Feature-level classifier accuracy for each feature type and mean across all feature types

	Ventral Stream				Dorsal Stream
	V1	V2v	V3v	LOC	V2d	V3d
Feat 1 (SF1)	0.791 (±0.0261)	0.722 (±0.0254)	0.686 (±0.0178)	0.516 (± 0.0051)	0.689 (±0.0237)	0.652 (±0.0195)
Feat 2 (Outline1)	0.864 (±0.0163)	0.675 (±0.0188)	0.609 (±0.0186)	0.550 (±0.0101)	0.777 (±0.0239)	0.618 (±0.0203)
Feat 3 (SF2)	0.773 (±0.0235)	0.679 (±0.0204)	0.658 (±0.0178)	0.526 (±0.0092)	0.685 (±0.0211)	0.653 (±0.0206)
Feat 4 (Outline2)	0.853 (±0.0160)	0.678 (±0.0223)	0.657 (±0.0176)	0.557 (±0.0062)	0.736 (±0.0240)	0.692 (±0.0280)
Overall mean acc.	0.820 (±0.0186)	0.688 (±0.0186)	0.653 (± 0.0121)	0.537 (±0.00266)	0.722 (±0.02)	0.653 (±0.0146)
95% CI around overall mean	[0.815, 0.825]	[0.682, 0.695]	[0.646, 0.659]	[0.530, 0.544]	[0.716, 0.728]	[0.647, 0.66]

In both ventral and dorsal streams, 1st 4 rows show mean (±SE) across subjects of the feature-level classifier accuracy for each feature type across subjects. Chance = 0.5. SF,  spatial frequency. Fifth row shows overall mean ± SE accuracy across all 4 feature-level classifiers  across subjects. Sixth row shows 95% CIs around the overall mean accuracy determined by within-subject bootstrap resampling with replacement over 10,000 iterations. All data were averaged over 2 sessions for each subject. Overall mean accuracy scores for each session in each subject were transformed into a log-likelihood ratio (log odds; chance = 0) and submitted to a 2-way repeated-measures ANOVA, with factors Scan Session (1, 2) and ROI (V1, V2v, V3v, V2d, V3d, LOC), revealing a main effect of ROI [F(5,35) = 83.14, P < 0.001, η² = 0.922], no significant effect of Scan Session [F(1,7) = 0.860, P = 0.39, η² = 0.109], and a nonsignificant interaction [F(5,35) = 2.367, P = 0.06, η² = 0.253]. Accuracy was lowest in region LOC, but a 1-sample t-test revealed that the log odds of LOC accuracy (collapsed over sessions) exceeded chance performance (mean = 0.1486, SD = 0.0303, t[7] = 13.89, P < 0.001).