For　some　professional　sports,　it　is　highly　required　to　supervise　and　analyze　the　athletics　pose　in　training　of　athletes.　Key　frame　extraction　from　training　videos　plays　a　key　role　to　facilitate　the　browse　of　sport　training　videos.　In　this　paper,　we　propose　a　deep　key　frame　extraction　method　for　analyzing　weightlifting　sport　training　videos.　To　alleviate　the　bias　from　complex　background,　Fully　Convolutional　Networks　(FCN)　is　employed　firstly　to　extract　the　region　of　interest　(ROI)　which　contains　mainly　the　athlete　and　barbell　for　a　more　precise　pose　estimation　of　frames.　Then　over　the　extracted　ROI,　Convolutional　Neural　Networks　(CNN)　are　leveraged　to　estimate　the　pose　probability　of　each　frame.　Finally,　a　variation　aware　key　frame　extraction　is　constructed　to　extract　the　key　frames　considering　neighboring　probability　difference　of　frames.　The　experimental　results　demonstrate　that　the　proposed　method　achieves　good　performance　in　key　frame　extraction　of　sport　videos,　and　significantly　outperforms　the　comparisons.　(C)　2018　Published　by　Elsevier　B.V.