Python library to work with geospatial raster and vector data for deep learning.
rsnet is designed to make it easier for the deep learning researchers to handle the remote sensing data.
- perform model prediction for large remote sensing image.
from rsnet.dataset import RasterSampleDataset
from torch.utils.data import DataLoader
from torchvision import transforms as T
tsf = T.Compose([
T.ToTensor(),
T.Normalize(mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225))
])
ds = RasterSampleDataset('example.tif',
win_size=512,
step_size=512,
pad_size=128,
band_index=(3, 2, 1),
transform=tsf)
# Deep learning model predict
loader = DataLoader(ds,
batch_size=1,
num_workers=0,
shuffle=False,
drop_last=False)
for img, xoff, yoff in loader:
with torch.no_grad():
result = model(img)- split large image into tile image.
from rsnet.converter import RasterDataSpliter
ds_spliter = RasterDataSpliter('example.tif',
win_size=512,
step_size=512)
ds_spliter.run('/path/to/output', progress=True)- Eval classification result
from rsnet.eval import eval_seg
pred_fname = '/path/to/pred.tif'
gt_fname = '/path/to/gt.tif'
ret_metrics = eval_seg(pred_fname,
gt_fname,
num_classes=5,
metrics=['IoU', 'Prec', 'Recall'])- Rasterize vector to raster
import rasterio as rio
from rsnet.converter import rasterize
vfile = '/path/to/vectorfile'
rfile = '/path/to/reference/rasterfile'
image = rasterize(vfile, rfile, property='id')
# write file
with rio.open(rfile, 'r') as src:
with rio.open('rasterized-results.tif',
'w',
driver='GTiff',
dtype=rio.uint8,
count=1,
width=src.width,
height=src.height,
transform=src.transform) as dst:
dst.write(image, indexes=1)