#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Aug 7 17:50:47 2019
@author: lferiani
"""
#%% import statements
import numpy as np
import pandas as pd
from numpy.fft import fft2, ifft2, fftshift
#%% constants
# dictionary to go from camera name to channel
# to be updated as we get more copies of the LoopBio rig
CAM2CH_DICT_legacy = {"22594549":'Ch1',
"22594548":'Ch2',
"22594546":'Ch3',
"22436248":'Ch4',
"22594559":'Ch5',
"22594547":'Ch6'}
CAM2CH_DICT = {"22956818":'Ch1', # Hydra01
"22956816":'Ch2',
"22956813":'Ch3',
"22956805":'Ch4',
"22956807":'Ch5',
"22956832":'Ch6',
"22956839":'Ch1', # Hydra02
"22956837":'Ch2',
"22956836":'Ch3',
"22956829":'Ch4',
"22956822":'Ch5',
"22956806":'Ch6',
"22956814":'Ch1', # Hydra03
"22956833":'Ch2',
"22956819":'Ch3',
"22956827":'Ch4',
"22956823":'Ch5',
"22956840":'Ch6',
"22956812":'Ch1', # Hydra04
"22956834":'Ch2',
"22956817":'Ch3',
"22956811":'Ch4',
"22956831":'Ch5',
"22956809":'Ch6',
"22594559":'Ch1', # Hydra05
"22594547":'Ch2',
"22594546":'Ch3',
"22436248":'Ch4',
"22594549":'Ch5',
"22594548":'Ch6'}
# this can't be a nice and simple dictionary because people may want to use
# this info in the other direction
CAM2CH_list = [('22956818', 'Ch1', 'Hydra01'), # Hydra01
('22956816', 'Ch2', 'Hydra01'),
('22956813', 'Ch3', 'Hydra01'),
('22956805', 'Ch4', 'Hydra01'),
('22956807', 'Ch5', 'Hydra01'),
('22956832', 'Ch6', 'Hydra01'),
('22956839', 'Ch1', 'Hydra02'), # Hydra02
('22956837', 'Ch2', 'Hydra02'),
('22956836', 'Ch3', 'Hydra02'),
('22956829', 'Ch4', 'Hydra02'),
('22956822', 'Ch5', 'Hydra02'),
('22956806', 'Ch6', 'Hydra02'),
('22956814', 'Ch1', 'Hydra03'), # Hydra03
('22956833', 'Ch2', 'Hydra03'),
('22956819', 'Ch3', 'Hydra03'),
('22956827', 'Ch4', 'Hydra03'),
('22956823', 'Ch5', 'Hydra03'),
('22956840', 'Ch6', 'Hydra03'),
('22956812', 'Ch1', 'Hydra04'), # Hydra04
('22956834', 'Ch2', 'Hydra04'),
('22956817', 'Ch3', 'Hydra04'),
('22956811', 'Ch4', 'Hydra04'),
('22956831', 'Ch5', 'Hydra04'),
('22956809', 'Ch6', 'Hydra04'),
('22594559', 'Ch1', 'Hydra05'), # Hydra05
('22594547', 'Ch2', 'Hydra05'),
('22594546', 'Ch3', 'Hydra05'),
('22436248', 'Ch4', 'Hydra05'),
('22594549', 'Ch5', 'Hydra05'),
('22594548', 'Ch6', 'Hydra05')]
CAM2CH_df = pd.DataFrame(CAM2CH_list,
columns=['camera_serial', 'channel', 'rig'])
# dictionaries to go from channel/(col, row) to well name.
# there will be many as it depends on total number of wells, upright/upsidedown,
# and in case of the 48wp how many wells in the fov
UPRIGHT_48WP_669999 = pd.DataFrame.from_dict({ ('Ch1',0):['A1','B1','C1'],
('Ch1',1):['A2','B2','C2'],
('Ch2',0):['D1','E1','F1'],
('Ch2',1):['D2','E2','F2'],
('Ch3',0):['A3','B3','C3'],
('Ch3',1):['A4','B4','C4'],
('Ch3',2):['A5','B5','C5'],
('Ch4',0):['D3','E3','F3'],
('Ch4',1):['D4','E4','F4'],
('Ch4',2):['D5','E5','F5'],
('Ch5',0):['A6','B6','C6'],
('Ch5',1):['A7','B7','C7'],
('Ch5',2):['A8','B8','C8'],
('Ch6',0):['D6','E6','F6'],
('Ch6',1):['D7','E7','F7'],
('Ch6',2):['D8','E8','F8']})
UPRIGHT_96WP = pd.DataFrame.from_dict({('Ch1',0):[ 'A1', 'B1', 'C1', 'D1'],
('Ch1',1):[ 'A2', 'B2', 'C2', 'D2'],
('Ch1',2):[ 'A3', 'B3', 'C3', 'D3'],
('Ch1',3):[ 'A4', 'B4', 'C4', 'D4'],
('Ch2',0):[ 'E1', 'F1', 'G1', 'H1'],
('Ch2',1):[ 'E2', 'F2', 'G2', 'H2'],
('Ch2',2):[ 'E3', 'F3', 'G3', 'H3'],
('Ch2',3):[ 'E4', 'F4', 'G4', 'H4'],
('Ch3',0):[ 'A5', 'B5', 'C5', 'D5'],
('Ch3',1):[ 'A6', 'B6', 'C6', 'D6'],
('Ch3',2):[ 'A7', 'B7', 'C7', 'D7'],
('Ch3',3):[ 'A8', 'B8', 'C8', 'D8'],
('Ch4',0):[ 'E5', 'F5', 'G5', 'H5'],
('Ch4',1):[ 'E6', 'F6', 'G6', 'H6'],
('Ch4',2):[ 'E7', 'F7', 'G7', 'H7'],
('Ch4',3):[ 'E8', 'F8', 'G8', 'H8'],
('Ch5',0):[ 'A9', 'B9', 'C9', 'D9'],
('Ch5',1):['A10','B10','C10','D10'],
('Ch5',2):['A11','B11','C11','D11'],
('Ch5',3):['A12','B12','C12','D12'],
('Ch6',0):[ 'E9', 'F9', 'G9', 'H9'],
('Ch6',1):['E10','F10','G10','H10'],
('Ch6',2):['E11','F11','G11','H11'],
('Ch6',3):['E12','F12','G12','H12']})
#%% functions
def get_mwp_map(total_n_wells, whichsideup):
"""
Given a total number of wells, and whether the multiwell plate
is upright or upside-down, returns a dataframe with the correct
channel/row/column -> well_name mapping
(this works on the Hydra imaging systems - by LoopBio Gmbh - used in Andre
Brown's lab)
"""
if total_n_wells==48 and whichsideup=='upright':
return UPRIGHT_48WP_669999
elif total_n_wells==96 and whichsideup=='upright':
return UPRIGHT_96WP
else:
raise ValueError('This case has not been coded yet. ' + \
'Please contact the devs or open a feature request on GitHub.')
def serial2rigchannel(camera_serial):
"""
Takes camera serial number, returns a (rig, channel) tuple
"""
out = CAM2CH_df[CAM2CH_df['camera_serial']==camera_serial]
if len(out) == 0:
raise ValueError('{} unknown as camera serial string'.format(camera_serial))
elif len(out) == 1:
return tuple(out[['rig','channel']].values[0])
else:
raise Exception('Multiple hits for {}. split_fov/helper.py corrupted?'.format(camera_serial))
def serial2channel(camera_serial):
"""
Takes camera serial number, returns the channel
"""
return serial2rigchannel(camera_serial)[1]
def parse_camera_serial(filename):
import re
regex = r"(?<=20\d{6}\_\d{6}\.)\d{8}"
camera_serial = re.findall(regex, str(filename).lower())[0]
return camera_serial
def calculate_bgnd_from_masked_fulldata(masked_image_file):
"""
- Opens the masked_image_file hdf5 file, reads the /full_data node and
creates a "background" by taking the maximum value of each pixel over time.
- Parses the file name to find a camera serial number
- reads the pixel/um ratio from the masked_image_file
"""
import numpy as np
from tierpsy.helper.params import read_unit_conversions
# read attributes of masked_image_file
_, (microns_per_pixel, xy_units) , is_light_background = read_unit_conversions(masked_image_file)
# get "background" and px2um
with pd.HDFStore(masked_image_file, 'r') as fid:
assert is_light_background, \
'MultiWell recognition is only available for brightfield at the moment'
img = np.max(fid.get_node('/full_data'), axis=0)
camera_serial = parse_camera_serial(masked_image_file)
return img, camera_serial, microns_per_pixel
def make_square_template(n_pxls=150, rel_width=0.8, blurring=0.1, dtype_out='float'):
import numpy as np
"""Function that creates a template that approximates a square well"""
n_pxls = int(np.round(n_pxls))
x = np.linspace(-0.5, 0.5, n_pxls)
y = np.linspace(-0.5, 0.5, n_pxls)
xx, yy = np.meshgrid(x, y, sparse=False, indexing='ij')
# inspired by Mark Shattuck's function to make a colloid's template
zz = (1 - np.tanh( (abs(xx)-rel_width/2)/blurring ))
zz = zz * (1-np.tanh( (abs(yy)-rel_width/2)/blurring ))
zz = zz/4
# add bright border
edge = int(0.05 * n_pxls)
zz[:edge,:] = 1
zz[-edge:,:] = 1
zz[:,:edge] = 1
zz[:,-edge:] = 1
if dtype_out == 'uint8':
zz *= 255
zz = zz.astype(np.uint8)
elif dtype_out == 'float':
pass
else:
raise ValueError("Only 'float' and 'uint8' are valid dtypes for this")
return zz
def was_fov_split(timeseries_data):
"""
Check if the FOV was split, looking at timeseries_data
"""
if 'well_name' not in timeseries_data.columns:
# for some weird reason, save_feats_stats is being called on an old
# featuresN file without calling save_timeseries_feats_table first
is_fov_split = False
else:
# timeseries_data has been updated and now has a well_name column
if len(set(timeseries_data['well_name']) - set(['n/a'])) > 0:
is_fov_split = True
# print('have to split fov by well')
else:
assert all(timeseries_data['well_name']=='n/a'), \
'Something is wrong with well naming - go check save_feats_stats'
is_fov_split = False
return is_fov_split
def naive_normalise(img):
m = img.min()
M = img.max()
return (img - m) / (M-m)
def fft_convolve2d(x,y):
""" 2D convolution, using FFT"""
fr = fft2(x)
fr2 = fft2(y)
cc = np.real(ifft2(fr*fr2))
cc = fftshift(cc)
return cc
def simulate_wells_lattice(img_shape, x_off, y_off, sp, nwells=None, template_shape='square'):
"""
Create mock fov by placing well templates onto a square lattice
Very simply uses the input parameters and range to define where the wells
will go, and then places the template in a padded canvas.
The canvas is then cut to be of img_shape again.
This simple approach works because the template is created to be exactly
spacing large, so templates do not overlap
"""
# convert fractions into integers
x_offset = int(x_off*img_shape[0])
y_offset = int(y_off*img_shape[0])
spacing = int(sp*img_shape[0])
# create a padded empty canvas
padding = img_shape[0]//2
padding_times_2 = padding*2
padded_shape = tuple(s+padding_times_2 for s in img_shape)
padded_canvas = np.zeros(padded_shape)
# determine where the wells wil go in the padded canvas
if nwells is not None:
r_wells = range(y_offset+padding,
y_offset+padding+nwells*spacing,
spacing)
c_wells = range(x_offset+padding,
x_offset+padding+nwells*spacing,
spacing)
else:
r_wells = range(y_offset+padding,
padding+img_shape[0],
spacing)
c_wells = range(x_offset+padding,
padding+img_shape[1],
spacing)
tmpl_pos_in_padded_canvas = [(r,c) for r in r_wells for c in c_wells]
# make the template for the wells
tmpl = make_square_template(n_pxls=spacing,
rel_width=0.7,
blurring=0.1,
dtype_out='float')
# invert
tmpl = 1-tmpl
# place wells onto canvas
ts = tmpl.shape[0]
for r,c in tmpl_pos_in_padded_canvas:
try:
padded_canvas[r-ts//2:r-(-ts//2),
c-ts//2:c-(-ts//2)] += tmpl
except Exception as e:
print(str(e))
import pdb
pdb.set_trace()
cutout_canvas = padded_canvas[padding:padding+img_shape[0],
padding:padding+img_shape[1]]
cutout_canvas = naive_normalise(cutout_canvas)
return cutout_canvas
if __name__ == '__main__':
# test that camera serials return the correct channel
serials_list = [line[0] for line in CAM2CH_list]
# serials_list.append('22594540') # this raise an exception as it does not exist
for serial in serials_list:
print('{} -> {}'.format(serial, serial2channel(serial)))
# that works as intended!
# let's now check that the camera name is parsed correctly I guess
from pathlib import Path
src_dir = Path('/Users/lferiani/Desktop/Data_FOVsplitter/evgeny/MaskedVideos/20190808')
masked_fnames = src_dir.rglob('*.hdf5')
for fname in masked_fnames:
camera_serial = parse_camera_serial(fname)
print(fname)
print(camera_serial)
print(serial2channel(camera_serial))
print(' ')
# this too works perfectly... but I saw wrong data was written in the masked videos
# so have to check what went wrong there
