multi-scale classification code

[zwh-2017]

Dear author, Could you release your multi-scale classification code? Thank you。

[amiltonwong]

Hi, @Yochengliu ,
The best result mentioned in your paper is multi-scale (s=3) version. When will the multi-scale (s=3) version be released? If it's not yet ready, could you give some hints on how to build it by extend your single-scale version?

THX!

[yuan-zm]

I think the only thing we should do is modify the rscnn_ssn_cls.py.
Try to use this.
`
self.SA_modules.append(
PointnetSAModuleMSG(
npoint=512,

            first_layer=True,
            radii=[0.1, 0.2, 0.4],
            nsamples=[16, 32, 128],
            mlps=[[input_channels, 64], [input_channels, 64], [input_channels, 64]],

            use_xyz=use_xyz,
            relation_prior=relation_prior
        )
    )

    c_out_0 = 64 * 3

    self.SA_modules.append(
        PointnetSAModuleMSG(
            npoint=128,
            radii=[0.2, 0.4, 0.8],
            nsamples=[32, 64, 128],
            mlps=[
                [c_out_0, 512],
                [c_out_0, 512],
                [c_out_0, 512],
            ],

            use_xyz=use_xyz,
            relation_prior=relation_prior
        )
    )
    c_out_1 = 512 * 3
    self.SA_modules.append(
        # global convolutional pooling
        PointnetSAModule(
            nsample = 128,
            mlp=[c_out_1, 1024], 
            
            use_xyz=use_xyz
        )
    )

`
Sadly, the result is not good. I don't get the result released by the author. Maybe the nsamples and radii are not good.