dl

Deep learning based operators

---

source

download_dl_model

 download_dl_model (models:str|list=None, path:str=None)

	Type	Default	Details
models	str | list	None	deep learning models to be downloaded, all available models by default
path	str	None	directory to save these models, inside installed Moraine package by default

Deep learning models have to be installed before use, download them by:

download_dl_model()

Downloading https://raw.githubusercontent.com/kanglcn/n2f/refs/heads/main/n2f.onnx to /work/projects/jinwook/radarlab/kangl/moraine/moraine/dl_model/n2f.onnx
File '/work/projects/jinwook/radarlab/kangl/moraine/moraine/dl_model/n2f.onnx' downloaded successfully.
Downloading https://raw.githubusercontent.com/kanglcn/n2f/refs/heads/main/n2fs3d.onnx to /work/projects/jinwook/radarlab/kangl/moraine/moraine/dl_model/n2fs3d.onnx
File '/work/projects/jinwook/radarlab/kangl/moraine/moraine/dl_model/n2fs3d.onnx' downloaded successfully.

---

source

n2f

 n2f (intf:numpy.ndarray, chunks:tuple=None, depths:tuple=(0, 0),
      model:str=None)

	Type	Default	Details
intf	ndarray		interferogram, 2d np.complex64 or cp.complex64
chunks	tuple	None	chunksize, intf.shape by default
depths	tuple	(0, 0)	width of the boundary
model	str	None	path to the model in onnx format, use the model comes with this package by default

Usage:

import zarr
import numpy as np
import holoviews as hv
hv.extension('bokeh')
hv.output(widget_location='bottom')
from bokeh.models import WheelZoomTool
from holoviews import opts

rslc_zarr = zarr.open('../CLI/raw/rslc.zarr/','r')
intf = rslc_zarr[:,:,7]*rslc_zarr[:,:,13].conj()

intf.shape

(2500, 1834)

filtered_intf = n2f(intf,depths=(10,10),model='../../../noise2fringe_ps_cos/model_saved/noise2fringe_model_depth_4_0712_5/n2f.onnx')

CPU times: user 4min 6s, sys: 43.2 s, total: 4min 49s
Wall time: 5.06 s

raw_intf_plot = mr.ras_plot(np.angle(intf))
filtered_intf_plot = mr.ras_plot(np.angle(filtered_intf))
(raw_intf_plot+filtered_intf_plot).opts(
    opts.Image(
        cmap='colorwheel',width=600, height=600, colorbar=True,invert_yaxis=True,
        default_tools=['pan',WheelZoomTool(zoom_on_axis=False),'save','reset','hover'],
        active_tools=['wheel_zoom'])
)

---

source

n2fs3d

 n2fs3d (adi:numpy.ndarray, intf:numpy.ndarray, chunks:tuple=None,
         depths:tuple=(0, 0), model:str=None)

	Type	Default	Details
adi	ndarray		amplitude dispersion index, 2d np.float32 or cp.float32
intf	ndarray		interferogram, 2d np.complex64 or cp.complex64
chunks	tuple	None	chunksize, intf.shape by default
depths	tuple	(0, 0)	width of the boundary
model	str	None	path to the model in onnx format, use the model comes with this package by default

Usage:

import zarr
import numpy as np
import holoviews as hv
hv.extension('bokeh')
hv.output(widget_location='bottom')
from bokeh.models import WheelZoomTool
from holoviews import opts

rslc_zarr = zarr.open('../CLI/raw/rslc.zarr/','r')
intf = rslc_zarr[:,:,7]*rslc_zarr[:,:,13].conj()
adi = zarr.open('../CLI/ps/adi.zarr/','r')[:]

filtered_intf = n2fs3d(adi,intf,depths=(200,200),model='../../../noise2fringe_ps_cos_adi/model_saved/noise2fringe_model_depth_4_0721_8/n2fs3d.onnx')

CPU times: user 3min 36s, sys: 1min 1s, total: 4min 37s
Wall time: 5.02 s

if is_cuda_available():
    intf_cp = cp.asarray(intf)
    adi_cp = cp.asarray(adi)
    filtered_intf_cp = n2fs3d(adi_cp,intf_cp)

CPU times: user 3.76 s, sys: 613 ms, total: 4.38 s
Wall time: 4.34 s