Large-Scale Coarse-to-Fine Object Retrieval Ontology and Deep Local Multitask Learning

Table 10

Recall in STL and MTL for fashion attributes.

Attribute	STL	MTL	Local MTL
lace	0.7049	0.4076	0.6185
knit	0.5051	0.3371	0.6606
denim	0.7567	0.6203	0.8244
chiffon	0.3390	0.1717	0.7538
dye	0.0	0.3027	0.3561
fur	0.3308	0.6875	0.6654
metallic	0.0	0.0	0.1573
sleeve	0.6018	0.0753	0.6876
sleeveless	0.6977	0.1315	0.6574
v-neck	0.3602	0.2025	0.5702
collar	0.1152	0.0827	0.3320
button	0.0524	0.0690	0.4952
zip	0.3048	0.2055	0.5173
bow	0.0	0.0331	0.3471
maxi	0.8345	0.7730	0.8560
shirt	0.7950	0.4042	0.8117
fit	0.3768	0.2464	0.7127
bodycon	0.5808	0.5234	0.7916
mini	0.3365	0.2647	0.3593
midi	0.2969	0.3099	0.4356
slim	0.4474	0.2	0.4330
summer	0.6172	0.0	0.5512
classic	0.5487	0.0070	0.7025
party	0.0329	0.0	0.4152
chic	0.0	0.0	0.1235
solid	0.0	0.0	0.1810
workout	0.5424	0.2768	0.6215
varsity	0.0	0.1890	0.4329
print	0.8592	0.5124	0.8521
floral	0.6521	0.4540	0.6264
striped	0.6935	0.5829	0.7505
dot	0.3925	0.4150	0.4935
linen	0.0	0.0455	0.4163
marled	0.4	0.2722	0.6250
leopard	0.0	0.2966	0.3118
Average	0.3764	0.2600	0.5470

Italics: attributes with fewer data than others.