Most　recent　single-view　3D　reconstruction　methods　still　utilize　3D　priors　or　semantic　priors,　while　not　making　use　of　the　projective　geometry　relation　between　2D　image　and　3D　mesh,　as　well　as　2D　image　priors.　In　the　absence　of　priors,　self-supervised　methods　can　be　effective　for　the　3D　reconstruction　of　images.　However,　without　the　ground　truth,　self-supervision　is　prone　to　converge　to　sub-optimal　states　and　is　ineffective　for　complex　objects.　To　address　this　problem,　in　this　paper,　we　propose　an　image　priors-based　method,　denoted　as　Model-Guided　Self-Supervised　3D　Mesh　Reconstruction　(MoG-SMR).　On　the　one　hand,　we　embed　the　NARM　combined　with　cross-scale　into　our　method　to　address　problems　such　as　occlusion　and　blurring.　On　the　other　hand,　we　make　use　of　external　priors　gained　from　other　images　to　improve　the　quality　of　3D　mesh　reconstruction.　Experimental　results　demonstrate　that　our　MoG-SMR　outperforms　state-of-The-Art　methods　in　terms　of　subjective　and　objective　quality　of　the　3D　mesh　reconstruction.　©　2024　ACM.