Table 4. 1-NN classification accuracy in the visual object recognition study using the DeCAF fully connected layer fc7.

fc7	Train on source No adapt.	Unsup. GFK [13]	DA labels: S		labels: S, T KEMA Kint	Train on target No adapt.
			OT-lab [15]	JDA [26]
lS		0	20	20	20
lT		0	0	$\widehat{y}$	3
C → A	84.5±1.5	87.8±2.1	92.1 ± 1.3	89.6±2.0	91.5±1.5	84.4±3.6
C → D	73.1±4.9	83.5±3.6	85.4±6.0	85.0±4.9	93.6 ± 3.1	92.2±1.9
A → C	72.0±1.7	80.2±1.9	87.2 ± 1.2	82.6±2.9	80.3±3.4	66.3±3.7
A → W	61.3±3.4	78.0±4.8	84.5±2.4	83.0±4.6	92.7 ± 2.5	88.1±3.8
W → C	68.9±3.0	75.1±2.5	83.7 ± 1.5	79.8±2.0	82.1±2.3	66.3±3.7
W → A	73.5±2.7	81.2±2.2	91.9 ± 1.4	90.9±1.2	91.6±1.3	84.4±3.6
D → A	74.6±3.9	85.4±2.1	92.9 ± 1.1	91.9±0.8	90.3±1.1	84.4±3.6
D → W	93.8±1.5	96.7±1.9	94.1 ± 3.4	97.0±1.5	91.0±3.5	88.1±3.8
Mean	75.2	83.5	88.9	87.49	89.1	81.7

C: Caltech, A: Amazon, D: DSLR, W: Webcam.

l_domain: number of labels per class.

$\widehat{y}$: predicted labels.