Incompatible image size with RandomGeoSampler

[hfangcat]

Discussed in #1773

^{Originally posted by hfangcat December 14, 2023}
The code example is here:

from torchgeo.datasets import stack_samples

from torch.utils.data import DataLoader

from torchgeo.samplers import RandomGeoSampler, RandomBatchGeoSampler
from torchgeo.datasets import ChesapeakeCVPR

root = '/scratch/local/cvpr_chesapeake_landcover/'
dataset_train = ChesapeakeCVPR(root, splits=['md-train'], layers=['naip-new', 'lc', 'nlcd'], download=False)

sampler_train = RandomGeoSampler(dataset_train, size=224, length=1)
loader_train = DataLoader(dataset_train, sampler=sampler_train, collate_fn=stack_samples)

for i, data in enumerate(loader_train):
    images = data["image"]
    targets = data["mask"]
    print(images.shape, targets.shape)
    break

Then it shows the shape of images and targets:

torch.Size([1, 4, 176, 176]) torch.Size([1, 2, 176, 176])

which is not compatible with 224: what I specified in the code sampler_train = RandomGeoSampler(dataset_train, size=224, length=1), can the community help me find out why it happens?

[dmeaux]

Since you're just looking at Maryland, have you tried using ChesapeakeMD since it inherits from Chesapeake, which inherits from RasterDataset? I would think that each state's own class would give you a consistent projected resolution.

[hfangcat]

Since you're just looking at Maryland, have you tried using ChesapeakeMD since it inherits from Chesapeake, which inherits from RasterDataset? I would think that each state's own class would give you a consistent projected resolution.

@dmeaux Thanks! Actually, I will use all datasets afterward for training, this is just an example to test the dataloader :D But I can try what you mentioned if it is really hard to fix

[adamjstewart]

Note that ChesapeakeMD and friends only have masks, not images. Only ChesapeakeCVPR has both images and masks.

[calebrob6]

Note, this could be solved if we create the concept of TileDatasets (being discussed in #1353).

[adamjstewart]

Or if we convert ChesapeakeCVPR to a RasterDataset.