The　applications　of　3D　mesh,　such　as　computer　vision,　molecular　biology,　and　medical　imaging,　have　received　sufficient　attention　recently.　Polygonal　meshes　are　a　significant　raw　representation　of　3D　shapes　thanks　to　their　compactness　and　flexibility.　Though　extensive　research　efforts　are　concentrating　on　how　to　represent　3D　shapes　well　based　on　voxels　and　point　clouds,　there　is　little　effort　on　meshes　due　to　their　irregularity.　To　tackle　the　challenge,　we　propose　a　new　mesh　representation　called　Heat　Diffusion　Mesh　(HDM)　in　this　paper.　It　represents　meshes　that　are　in　non-Euclidean　domains　as　a　space-time　signal　by　heat　diffusion.　Experiments　of　mesh　classification　and　segmentation　show　the　effectiveness　of　the　proposed　mesh　representation.　©　2022　ACM.