batch dimension follows the band dimension? and missing metadata
import datasets
ds = datasets.load_dataset("satellogic/EarthView", streaming=True)
for sample in ds['train']:
print(sample)
break
test = sample['1m']
import numpy as np
np.array(test).shape
import numpy as np
np.array(test).shape
(3, 369, 64, 64)
It looks like the batch dimension follows the band dimension? Can documentation be added to the model card and to the description and info fields in the dataset itself? It also looks like the metadata field doesn't contain the full STAC metadata referenced in the article https://satellogic.com/2024/05/01/satellogic-open-source-release-a-large-dataset-of-high-resolution-imagery-for-ai-model-training/
{'bounds': [-82.02613963067623,
29.6093376452098,
-82.02547288964837,
29.609920270784258],
'epsg': 'EPSG:32617',
'siteID': 'OSBS',
'timestamp': ['2018-01-01T00:00:00',
'2019-01-01T00:00:00',
'2021-01-01T00:00:00']}
Dear @rbaveryw, thank you for your interest in our dataset. We are updating the README file as we speak. Regarding the STAC format, Satellogic STAC catalog has larger tiles than the ones we are sharing. The images and metadata we share here do not follow STAC format. As you can see, the shape you printed has four dimensions, (# of revisits, # of bands, height, width). The metadata corresponding to such sample includes a list of three timestamps as you show. The sample you downloaded corresponds to a NEON sample.