For ILSVRC 2012, we have found a consistent improvement of about 0.2%0.2\% absolute both for top-11 error Our highest quality version of Inception-v3 reaches 21.2%21.2\%, top-11 and 5.6%5.6\% top-5 error for based on denser networks: our model outperforms the results of He et al [6] —— cutting the top-55 (top