Unimodal#

[1]:

import sys
sys.path.append('../')

[2]:

import matplotlib.pyplot as plt
import torch
import numpy as np
import cv2
from utils import circle_size, colors, toImg, preprocess
from model import loadFan#, predict_landmarks
from torchgeometry.contrib import spatial_soft_argmax2d
from glob import glob
import tifffile as tiff

[3]:

def load_imgs(path):
    #get files with glob
    files = glob(f"{path}*")
    #sort files by number
    files = sorted(files, key=lambda x: int(x.split('/')[-1].split('.')[0]))
    #load images
    if files[0].endswith('.tif'):
        imgs = [tiff.imread(f) for f in files]
    else:
        imgs = [cv2.cvtColor(cv2.imread(f), cv2.COLOR_BGR2RGB) for f in files]
    return imgs

[4]:

#You can download the dataset from here: https://figshare.com/projects/ELD/167318

[5]:

#set your path here
PATH = "../../marcoAnalysis/"

MOB HE imgs#

[6]:

inpath = f'{PATH}MOB_eggplant/'

[7]:

#load imgs and move to tensor
mob_he = load_imgs(inpath)
image = torch.stack([preprocess(img) for img in mob_he])

You can can either train a model with:

python train.py --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/MOB_eggplant/ --npts 14 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --model unimodal

Or download a test model at: https://figshare.com/projects/ELD/167318

[8]:

#load model
fan = loadFan(npoints=14,n_channels=3,path_to_model="../models/unimodal/MOB_HE/model_110.fan.pth")

[9]:

def predict_landmarks(fan, image):
    with torch.no_grad():
        #img = sample['Im']

        fan.eval()

        with torch.no_grad():
            pts = 4 * spatial_soft_argmax2d(fan(image.cuda()), False)
    return pts

[10]:

#predict landmarks
pts = predict_landmarks(fan, image)
#combine landmarks and image
np_img = toImg(image.cuda()[:,:3], pts, 128)

[11]:

fig, axs = plt.subplots(3, 4, figsize=(15, 10))  # adjust the size as needed
axs = axs.ravel()

for i in range(12):
    img = np_img[i]
    axs[i].imshow(img)
    axs[i].set_title(f"Image {i+1}")
    axs[i].axis('off')  # to hide the axis

plt.tight_layout()
plt.show()
../_images/notebooks_unimodal_13_0.png

MOB 3 gene data#

[12]:

inpath = f'{PATH}MOB_3d/'

[13]:

#load imgs and move to tensor
mob_3d_imgs = load_imgs(inpath)
image = torch.stack([preprocess(img) for img in mob_3d_imgs])

You can can either train a model with:

python train.py --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/MOB_3d/ --npts 14 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --model unimodal

Or download a test model at: https://figshare.com/projects/ELD/167318

[14]:

#load model
fan = loadFan(npoints=14,n_channels=3,path_to_model="../models/unimodal/MOB_3d/model_120.fan.pth")

[15]:

#predict landmarks
pts = predict_landmarks(fan, image)
#combine landmarks and image
np_img = toImg(image.cuda()[:,:3], pts, 128)

[16]:

fig, axs = plt.subplots(3, 4, figsize=(15, 10))  # adjust the size as needed
axs = axs.ravel()

for i in range(12):
    img = np_img[i]
    axs[i].imshow(img)
    axs[i].set_title(f"Image {i+1}")
    axs[i].axis('off')  # to hide the axis

plt.tight_layout()
plt.show()
../_images/notebooks_unimodal_20_0.png

MOB 100 genes#

[17]:

inpath = f'{PATH}MOB_100d/'

[18]:

#load imgs and move to tensor
mob_100d_imgs = load_imgs(inpath)
image = torch.stack([preprocess(img) for img in mob_100d_imgs])

You can can either train a model with:

python train.py --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/MOB_100d/ --npts 14 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --model unimodal

Or download a test model at: https://figshare.com/projects/ELD/167318

[19]:

#load model
fan = loadFan(npoints=14,n_channels=100,path_to_model="../models/unimodal/MOB_100d/model_44.fan.pth")

[20]:

#predict landmarks
pts = predict_landmarks(fan, image)
#combine landmarks and image
np_img = toImg(image.cuda()[:,:3], pts, 128)

[21]:

fig, axs = plt.subplots(3, 4, figsize=(15, 10))  # adjust the size as needed
axs = axs.ravel()

for i in range(12):
    img = np_img[i]
    axs[i].imshow(img)
    axs[i].set_title(f"Image {i+1}")
    axs[i].axis('off')  # to hide the axis

plt.tight_layout()
plt.show()
../_images/notebooks_unimodal_27_0.png

ISS data#

[22]:

inpath = f'{PATH}paulo/'

[23]:

#load imgs and move to tensor
iss_imgs = load_imgs(inpath)
image = torch.stack([preprocess(img) for img in iss_imgs])

You can can either train a model with:

python train.py --elastic_sigma 5 --cuda 1 --port 9006 --data_path ../marcoAnalysis/paulo/ --npts 40 --o scratch --elastic True --step_size 5 --ws 0 --gamma 0.9 --angle 0.1 --model unimodal

Or download a test model at: https://figshare.com/projects/ELD/167318

[24]:

#load model
fan = loadFan(npoints=30,n_channels=3,path_to_model="../models/unimodal/iss/model_103.fan.pth")

[25]:

#predict landmarks
pts = predict_landmarks(fan, image)
#combine landmarks and image
np_img = toImg(image.cuda()[:,:3], pts, 128)

[26]:

fig, axs = plt.subplots(1,3, figsize=(15, 10))  # adjust the size as needed
axs = axs.ravel()

for i in range(3):
    img = np_img[i]
    axs[i].imshow(img)
    axs[i].set_title(f"Image {i+1}")
    axs[i].axis('off')  # to hide the axis

plt.tight_layout()
plt.show()
../_images/notebooks_unimodal_34_0.png 
The Kernel crashed while executing code in the the current cell or a previous cell. Please review the code in the cell(s) to identify a possible cause of the failure. Click <a href='https://aka.ms/vscodeJupyterKernelCrash'>here</a> for more info. View Jupyter <a href='command:jupyter.viewOutput'>log</a> for further details.