Table 4. Summary of MAP scores attained by our system in the public LifeCLEF 2014 Bird Identification Task (Goëau et al., 2014).

The first column lists scores attained locally in our two-fold lifeclef2014 split. The second column lists scores evaluated officially, using a classifier(s) trained across the entire training set.

System variant submitted	Cross-validated MAP (%)	Final official MAP (%)
melspec-kfl3-ms, noise red., binary relevance	30.56	36.9
Average from 12 single-layer models	32.73	38.9
melspec-kfl4pl8kfl4-ms, noise red., binary relevance	35.31	42.9
Average from 16 single- and double-layer models	35.07	41.4