How are the edge_index in MNISTSuperpixel Dataset computed?

[domalbert]

I was trying to use to transfer the MNIST dataset into superpixel representation using ToSlic and afterwards approximate the edge_index using KNNGraph.

import torch
import torch_geometric
from torchvision.datasets import MNIST
import torchvision.transforms as T
from torch_geometric.transforms import ToSLIC
from torch_geometric.transforms.knn_graph import KNNGraph
from torch_geometric.transforms.radius_graph import RadiusGraph

transform = T.Compose([T.ToTensor(),ToSLIC(n_segments=75),KNNGraph()])
dataset = MNIST('/tmp/MNIST', download=True, transform=transform)
train_set_x = dataset.data.numpy()#.reshape(1, 28,28, -1)
print(train_set_x.shape)

I visualized the graph representation for both MNIST and MNISTSuperpixel, however with strong differences in the result. For my custom adaption for MNIST I receive

while for inbuild Superpixel I obtain:

Can you give me guidance of how the edge_index is calculated within the MNISTSuperpixel representation?

If you need additional code let me know

[rusty1s]

I sadly don't have the SLIC parameters that were used to create the MNISTSuperpixelDataset. This dataset was pre-processed from the MoNet authors, I am sorry :(

[domalbert]

Okey. Thank you for your reply :-)