The correct identification of $b$-initiated jets against $c$-initiated, hadronic tau-initiated and $u$-, $d$-, $s$-initiated jets is primordial for ATLAS Collaboration searches and precision measurements e.g. $t\bar{t}$ and single-top measurements, supersymmetry, Higgs boson. In consequence, the ATLAS Collaboration has developed several so-called b-tagging algorithms to discriminate $b$-jets against other flavour jets. Expected performance is evaluated on Monte Carlo events, but due to some inaccuracies in the simulation, it might differ from the performance in real data events. These algorithms are calibrated to account for differences between data and simulated events in the $b$-tagging efficiency for $b$-jets and mis-tagging efficiency in $c$- and light-jets. The $b$-jet efficiency calibration is performed on two-lepton $t\bar{t}$ events in order to exploit the large purity of this final state in $b$-jets. The $c$-jet mis-tag rate is measured on one-lepton $t\bar{t}$ events, which allow the presence of an hadronically decaying $W$-jet enriched in $c$-jets. Both calibrations adopt a combinatorial likelihood approach to retrieve the efficiency of the jet for a given flavour. Finally, due to the large light-jet rejection of $b$-tagging algorithms, the calibration of light-jets mis-tag rate is performed on a $Z$+jet sample since $t\bar{t}$ doesn't allow for a pure enough light-jet sample. This proceeding presents an overview of the DL1r $b$-tagging algorithm and the calibration techniques explained above. Results in the form of data - Monte Carlo scale-factors are obtained for the $b$-jet efficiency and $c$- and light-jet mis-tag rate with 140 fb$^{-1}$ of data recorded by the ATLAS detector from 2015 to 2018