Image-text　retrieval　is　a　challenging　task　due　to　image　and　text　are　heterogeneous　cross-modal　data,　which　possess　semantic　gap.　The　key　issue　of　image-text　retrieval　is　how　to　learn　a　common　feature　space　while　semantic　correspondence　between　image　and　text　remains.　Some　existing　works　extract　region　feature　in　image　and　word　feature　in　text　to　implement　cross-modal　alignment　between　local　elements,　the　other　works　integrate　relation-aware　information　to　local　elements　to　compute　cross-modal　similarity,　while　these　methods　not　utilize　the　semantic　information　in　different　semantic-level.　In　order　to　address　this　issue,　we　propose　a　Bottom-up　Progressive　Semantic　Alignment　(BPSA)　network,　in　which　precise　fine-grained　alignment　is　carried　out　on　diverse　semantic-levels　progressively.　Specifically,　the　feature　of　the　cross-modal　data　are　extracted　from　bottom　element　to　local-group,　and　global-representation　by　graph　convolution　and　attention　mechanism.　We　conduct　extensive　experiments　on　Flickr30K　and　MS-COCO　datasets,　compared　with　the　related　state-of-the-art　methods.　The　results　show　that　our　network　achieves　competitive　performance.　©　2021,　Springer　Nature　Switzerland　AG.