NA60 is a fixed-target experiment at the CERN SPS which measured dimuon production in nucleus-nucleus and proton-nucleus collisions Thanks to the use of a radiation-tolerant silicon vertex telescope, placed immediately downstream of the target, the experiment has been able to collect a sample of muon pairs of unprecedented quality in heavy ion experiments. This telescope tracks the produced charged particles and allows us to select, among them, the two muons which gave the trigger. This matching procedure improves the dimuon mass resolution from around 80~MeV to around 20~MeV at the $\phi$ mass. Furthermore, the dipole field in the target region significantly increases the acceptance of low $p_{\rm T}$ and low mass dimuons. This allows NA60 to do a high quality measurement of $\phi$ yields and $p_{\rm T}$ distributions. This talk presents such measurements, done in In-In collisions at 158~GeV per nucleon, in 2003. The results, presented as a function of centrality, were studied against several possible sources of systematic effects (fit range, rapidity window, acceptance cuts etc.) and proved to be fairly stable. We show that the inverse $m_{\rm T}$ slope measured in In-In collisions, in the $\phi\to\mu\mu$ decay channel, increases with the number of nucleons participating in the collisions, rather than following a flat trend as seen in the NA50 data collected in the same decay channel but restricted to high $p_{\rm T}$ values. We also show that our measurements seem to agree with the values previously measured by NA49, using $\phi\to {\rm KK}$ decays, in Pb-Pb and other collision systems.