Neural　implicit　representations　are　highly　effective　for　single-view　3D　reconstruction　(SVR).　It　represents　3D　shapes　as　neural　fields　and　conditions　shape　prediction　on　input　image　features.　Image　features　can　be　less　effective　when　significant　variations　of　occlusions,　views,　and　appearances　exist　from　the　image.　To　learn　more　robust　features,　we　design　a　new　feature　encoding　scheme　that　works　in　both　image　and　shape　space.　Specifically,　we　present　a　geometry-aware　2D　convolutional　kernel　to　learn　image　appearance　and　view　information　along　with　geometric　relations.　The　convolutional　kernel　operates　at　the　2D　projections　of　a　point-based　3D　geometric　structure,　called　spatial　pattern.　Furthermore,　to　enable　the　network　to　discover　adaptive　spatial　patterns　that　capture　non-local　contexts,　the　kernel　is　devised　to　be　deformable　and　exploited　by　a　spatial　pattern　generator.　Experimental　results　on　both　synthetic　and　real　datasets　demonstrate　the　superiority　of　the　proposed　method.　©　The　Author(s),　under　exclusive　license　to　Springer　Nature　Switzerland　AG　2023.