Photo Aesthetics Ranking Network with Attributes and Content Adaptation

A Aesthetics & Attribute Database (AADB) Fusing Attributes and Content for Aesthetics Ranking Demo, code and model can be download through project webpage http://www.ics.uci.edu/~skong2/aesthetics.html References: [8] He, K., Zhang, X., Ren, S., Sun, J., ECCV, 2014 [15] Lu, X., Lin, Z., Jin, H., Yang, J., Wang, J., IEEE Trans. on Multimedia, 2015 [16] Lu, X., Lin, Z., Jin, H., Yang, J., Wang, J.Z., ACMMM, 2014 [17] Lu, X., Lin, Z., Shen, X., Mech, R., Wang, J.Z., ICCV, 2015 [23] Murray, N., Marchesotti, L., Perronnin, F., CVPR, 2012 Acknowledgements: This work was supported by Adobe gift fund, NSF grants DBI-1262547 and IIS1253538. Experimental Results We use Spearman's rho rank correlation ( ) to measure ranking performance . By thresholding the rating scores, we achieve state-of-the-art classification accuracy on AVA despite never training with a classification loss. We first train a simple model with Euclidean loss for numerical rating of photo aesthetics (a) fine-tuning with rank loss Based on the regression net, we apply rank loss to fine-tune the network