File size: 6,098 Bytes
a639402 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 | import torch
import wandb
import time
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
import pandas as pd
import pickle
import os
from torchvision.transforms import ToTensor, ToPILImage
from tqdm import tqdm
import random
from math import sqrt
def dummy(lists):
new = []
for i in lists:
new.append(i.item())
return new
def display(subject, ordered_t,count, num_images = 5):
def l1_dist(x, y):
return torch.sum(x - y).item()
def l2_dist(x, y):
return sqrt(torch.sum((x - y) ** 2).item())
# sort by distance to the subject
ordered_t = sorted(ordered_t, key=lambda elem: l2_dist(subject[0], elem[0]))
subject_repeated = [subject for _ in range(num_images)]
nearest_10_images = ordered_t[:num_images]
farthest_10_images = ordered_t[-num_images:]
def make_panel(list_of_images):
images = [image[1] for image in list_of_images]
labels = [image[2].cpu().numpy() for image in list_of_images]
filenames = [image[3] for image in list_of_images]
panel = torch.cat(images, dim=2)
panel_pil = ToPILImage().__call__(panel)
return panel_pil,labels, filenames
panel_of_subject,labels_sub, filenames = make_panel(subject_repeated)
#labels_sub = labels_sub.cpu().numpy()
labels_sub = labels_sub[0]
filename_sub = filenames[0]
#print('Input:',labels_sub)
panel_of_nearest_10,labels_near,_ = make_panel(nearest_10_images)
labels_near = dummy(labels_near)
#print('NN :',labels_near)
panel_of_farthest_10,labels_far,_ = make_panel(farthest_10_images)
#_img = np.concatenate((panel_of_subject, panel_of_nearest_10, panel_of_farthest_10), axis=0)
#plt.title(str(labels_sub) + str(labels_near))
#plt.imshow(_img)
#plt.savefig('D:\\CalcSupContrastExperiments\\nn\\'+'image'+str(count) +'.png',dpi=200)
#plt.close()
return labels_sub, labels_near, filename_sub
class Nearest_Neigbours:
def __init__(self, device, model, dataloaders, target_names, output_folders, name):
self.device = device
self.model = model
self.dataloaders = dataloaders
self.target_names = target_names
self.output_folders = output_folders
self.name = name
def extract(self):
self.model.eval()
minibatches_t = []
with torch.no_grad():
for i, (inputs, labels, img_name) in enumerate(self.dataloaders):
inputs, labels = inputs.to(self.device), labels.long().to(self.device)
features, feat = self.model(inputs)
#features = features.view(features.size(0), -1)
print(features.shape)
i_t = inputs.detach().cpu().unbind(0)
e_t = features.detach().cpu().unbind(0)
sublist_t = [elem_t for elem_t in zip(e_t, i_t, labels, img_name)]
minibatches_t.append(sublist_t)
ordered_t = []
for minibatch_t in minibatches_t:
while minibatch_t:
ordered_t.append(minibatch_t.pop())
with open(self.output_folders + 'ordered_features_'+self.name , "wb") as fp: #Pickling
pickle.dump(ordered_t, fp)
#with open(self.output_folders + 'labels_t'+self.name , "wb") as fp: #Pickling
# pickle.dump(labels_t, fp)
return ordered_t
def main_loop(self):
'''
if os.path.exists(self.output_folders + 'ordered_features_train'):
with open(self.output_folders + 'ordered_features_train', "rb") as fp: # Unpickling
ordered_t = pickle.load(fp)
#with open(self.output_folders + 'labels_t'+self.name, "rb") as fp: # Unpickling
# labels_t = pickle.load(fp)
print('')
else:
ordered_t = self.extract()
'''
if os.path.exists(self.output_folders + 'ordered_features_train_roi'):
with open(self.output_folders + 'ordered_features_train_roi', "rb") as fp: # Unpickling
ordered_t_roi = pickle.load(fp)
#with open(self.output_folders + 'labels_t'+self.name, "rb") as fp: # Unpickling
# labels_t = pickle.load(fp)
print('')
else:
ordered_t_roi = self.extract()
#encoded_features = torch.stack(encoded_features, dim=0)
num_features = len(ordered_t_roi)
inputs=[]
nns1 = []
nns2 = []
nns3 = []
nns4 = []
nns5 = []
filess =[]
encoded_representations = []
count = 0
for i in range(num_features):
count+=1
# pick a random image
print(count)
subject = ordered_t_roi[i]
labels_sub, labels_near, filename_sub = display(subject, ordered_t_roi, count, num_images = 5)
extract_features = ordered_t_roi[i][0]
#save_folder = os.makedirs(self.output_folders + 'extracted_features')
pth_filename = os.path.splitext(os.path.basename(filename_sub))[0]
pth_filename = self.output_folders + 'extracted_features/' + pth_filename +'.pth'
torch.save(extract_features,pth_filename)
filess.append(filename_sub)
inputs.append(labels_sub)
nns1.append(labels_near[0])
nns2.append(labels_near[1])
nns3.append(labels_near[2])
nns4.append(labels_near[3])
nns5.append(labels_near[4])
encoded_representations.append(pth_filename)
dataframe = pd.DataFrame({'Filename':filess,'Input':inputs,'NN1':nns1,'NN2':nns2,'NN3':nns3,'NN4':nns4,'NN5':nns5, 'Features':encoded_representations})
dataframe.to_csv(self.output_folders+'New' +self.name+'_nn.csv', index = False)
return |