Big Transfer Learning for Fine Art Classification

Table 2

State-of-the-art results for artist and style categorization on the Painting-91 dataset.

Reference	Method	Painting-91
		Artist	Style	Average
[37]	LDCF (learned from gram)	64.32	78.27	71.30
[39]	Cross-layer correlation	70.65	78.13	74.39
[34]	Structure selection	71.27	79.23	75.25
BiT-S (ours)	Big transfer learning	61.23	75.64	68.44
BiT-M (ours)	Big transfer learning	72.07	79.60	75.84

The bold values show that the proposed BiT-M model achieves state-of-the-art performance compared with previous work and BiT-S model.