6ddf825a4aa1bcf3a35bba8d36bc433fe5ebaa39,torchsample/utils.py,,th_nearest_interp_3d,#Any#Any#,285

Before Change

    
    // repeat coords along channel dim if not given that way
    if coords.dim() == 2:
        coords = coords.unsqueeze(0).repeat(input.size(0),1,1)

    // take clamp of coords so they"re in the image bounds
    xc = torch.clamp(coords[:,:,0], 0, input.size(1)-1)
    yc = torch.clamp(coords[:,:,1], 0, input.size(2)-1)
    zc = torch.clamp(coords[:,:,2], 0, input.size(3)-1)

    // round to nearest coordinate
    coords = torch.stack([xc.round().long(),
                          yc.round().long(),
                          zc.round().long()], 2)

    // gather image values at coordinates
    mapped_vals = torch.stack([th_gather_nd(input[i], coords[i])
                    for i in range(input.size(0))], 0)

    return mapped_vals.view_as(input)

After Change

    coords[:,1] = torch.clamp(coords[:,1], 0, input.size(2)-1).round()
    coords[:,2] = torch.clamp(coords[:,2], 0, input.size(3)-1).round()

    stride = torch.LongTensor(input.stride())[1:].float()
    idx = coords.mv(stride)

    input_flat = th_flatten(input)

Instances