forked from M-Labs/artiq
52 lines
1.5 KiB
Python
52 lines
1.5 KiB
Python
import numpy as np
|
|
from numba import jit
|
|
from scipy.optimize import least_squares
|
|
import logging
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
@jit(nopython=True)
|
|
def compute_gaussian(r, img_w, img_h,
|
|
gaussian_w, gaussian_h,
|
|
gaussian_cx, gaussian_cy):
|
|
for y in range(img_h):
|
|
for x in range(img_w):
|
|
ds = ((gaussian_cx-x)/gaussian_w)**2
|
|
ds += ((gaussian_cy-y)/gaussian_h)**2
|
|
r[x, y] = np.exp(-ds/2)
|
|
|
|
|
|
def fit(data, get_dataset):
|
|
img_w, img_h = data.shape
|
|
def err(parameters):
|
|
r = np.empty((img_w, img_h))
|
|
compute_gaussian(r, img_w, img_h, *parameters)
|
|
r -= data
|
|
return r.ravel()
|
|
guess = [
|
|
get_dataset("rexec_demo.gaussian_w", 12),
|
|
get_dataset("rexec_demo.gaussian_h", 15),
|
|
get_dataset("rexec_demo.gaussian_cx", img_w/2),
|
|
get_dataset("rexec_demo.gaussian_cy", img_h/2)
|
|
]
|
|
res = least_squares(err, guess)
|
|
return res.x
|
|
|
|
|
|
def get_and_fit():
|
|
if "dataset_db" in globals():
|
|
logger.info("using dataset DB for Gaussian fit guess")
|
|
def get_dataset(name, default):
|
|
try:
|
|
return dataset_db.get(name)
|
|
except KeyError:
|
|
return default
|
|
else:
|
|
logger.info("using defaults for Gaussian fit guess")
|
|
def get_dataset(name, default):
|
|
return default
|
|
get_dataset = lambda name, default: default
|
|
return fit(controller_driver.get_picture(), get_dataset)
|