Using deep narrow-band and broad-band imaging , we identify 401 z \approx 0.40 and 249 z \approx 0.49 H \alpha line-emitting galaxies in the Subaru Deep Field . Compared to other H \alpha surveys at similar redshifts , our samples are unique since they probe lower H \alpha luminosities , are augmented with multi-wavelength ( rest-frame 1000Å–1.5 \mu m ) coverage , and a large fraction ( 20 % ) of our samples has already been spectroscopically confirmed . Our spectra allow us to measure the Balmer decrement for nearly 60 galaxies with H \beta detected above 5 \sigma . The Balmer decrements indicate an average extinction of A ( { H } \alpha ) = 0.7 ^ { +1.4 } _ { -0.7 } mag . We find that the Balmer decrement systematically increases with higher H \alpha luminosities and with larger stellar masses , in agreement with previous studies with sparser samples . We find that the SFRs estimated from modeling the spectral energy distribution ( SED ) is reliable—we derived an “ intrinsic ” H \alpha luminosity which is then reddened assuming the color excess from SED modeling . The SED-predicted H \alpha luminosity agrees with H \alpha narrow-band measurements over 3 dex ( rms of 0.25 dex ) . We then use the SED SFRs to test different statistically-based dust corrections for H \alpha and find that adopting one magnitude of extinction is inappropriate : galaxies with lower luminosities are less reddened . We find that the luminosity-dependent dust correction of Hopkins et al . yields consistent results over 3 dex ( rms of 0.3 dex ) . Our comparisons are only possible by assuming that stellar reddening is roughly half of nebular reddening . The strong correspondence argue that with SED modeling , we can derive reliable intrinsic SFRs even in the absence of H \alpha measurements at z \sim 0.5 .