I have a question in the unpooling layer, parent.feat = parent.feat + point.feat[inverse]; here you return the parent, but what about the sparse_conv_feat; it would only have the original parent.feat rather than the parent.feat + point.feat[inverse]; For example after the unpooling; the first operation would be spconv.SubMConv3d wouldn't it run this rather than performing a replace feature parent.sparse_conv_feat = parent.sparse_conv_feat.replace_feature(parent.feat)?
I have a question in the unpooling layer,
parent.feat = parent.feat + point.feat[inverse]; here you return the parent, but what about thesparse_conv_feat; it would only have the original parent.feat rather than theparent.feat + point.feat[inverse]; For example after the unpooling; the first operation would bespconv.SubMConv3dwouldn't it run this rather than performing a replace featureparent.sparse_conv_feat = parent.sparse_conv_feat.replace_feature(parent.feat)?