Comments (5)
Hi,
Sorry for the late reply. I was busy with other projects.
I have uploaded the code for evaluation. (Link : https://github.com/SharifAmit/RVGAN/blob/master/eval.py)
If you find any problem with running the codes or results , please open an issue.
Thanks!
from rvgan.
from rvgan.
The printed confusion matrix should be 2-by-2. However it looks 3-by-3 in your attached image.
Did you change anything in the eval file ?
In you image it says the error is in 179 line as given below, however in the original eval.py file it is line 188.
tn, fp, fn, tp = confusion.ravel()
Also can you print the following before the above line ?
print(np.unique(y_true))
print(np.unique(y_pred))
It should give you the same number output for both the lines.
If it doesn't I believe either your y_true or y_pred is not int or float type. So its considering 1 and 1.0 as different values.
from rvgan.
Please open the issue, if your problem still persists.
Thanks
from rvgan.
I had the same issue of a 3x3 confusion matrix but it because had to use jaccard_score instead of jaccard_similarity_score because of sklearn version issues and jaccard_score doesn't have a normalize parameter. So y_true has values 0 an 255 while y_pred is 0 an 1 meaning there is 3 values all up. If add the line y_true[y_true==255]=1 it normalizes y_true and gives you a 2x2 confusion matrix instead of 3x3. I have the entire eval code below:
import glob
import os
import time
import argparse
import numpy as np
from PIL import Image
from libtiff import TIFF
import tensorflow as tf
import cv2
import keras
from tensorflow.keras.optimizers import Adam
from keras.models import Model
import keras.backend as K
from src.model import coarse_generator,fine_generator
from skimage.metrics import structural_similarity as ssim
from sklearn.metrics import jaccard_score
#from sklearn.metrics import jaccard_similarity_score
from sklearn.metrics import confusion_matrix,precision_recall_curve,f1_score,roc_auc_score,auc,recall_score, auc,roc_curve
global g_local_model
global g_global_model
def normalize_pred(img,mask):
img = np.reshape(img,[1,128,128,3])
mask = np.reshape(mask,[1,128,128,1])
img_coarse = tf.image.resize(img, (64,64), method=tf.image.ResizeMethod.LANCZOS3)
img_coarse = (img_coarse - 127.5) / 127.5
img_coarse = np.array(img_coarse)
mask_coarse = tf.image.resize(mask, (64,64), method=tf.image.ResizeMethod.LANCZOS3)
mask_coarse = (mask_coarse - 127.5) / 127.5
mask_coarse = np.array(mask_coarse)
_,x_global = g_global_model.predict([img_coarse,mask_coarse])
#print('uuu',x_global.shape, np.min(x_global),np.max(x_global)) # (1, 64, 64, 128) -4.0635333 16.58276
img = (img - 127.5) / 127.5
mask = (mask - 127.5) / 127.5
X_fakeB = g_local_model.predict([img,mask,x_global])
X_fakeB = (X_fakeB + 1) /2.0
X_fakeB = cv2.normalize(X_fakeB, None, alpha = 0, beta = 255, norm_type = cv2.NORM_MINMAX, dtype = cv2.CV_32F)
pred_img = X_fakeB[:,:,:,0]
return np.asarray(pred_img,dtype=np.float32)
def strided_crop(img, mask, img_h,img_w,height, width,stride=1):
full_prob = np.zeros((img_h, img_w),dtype=np.float32)
full_sum = np.ones((img_h, img_w),dtype=np.float32)
max_x = int(((img.shape[0]-height)/stride)+1)
#print("max_x:",max_x)
max_y = int(((img.shape[1]-width)/stride)+1)
#print("max_y:",max_y)
max_crops = (max_x)*(max_y)
i = 0
for h in range(max_x):
for w in range(max_y):
crop_img_arr = img[h * stride:(h * stride) + height,w * stride:(w * stride) + width]
crop_mask_arr = mask[h * stride:(h * stride) + height,w * stride:(w * stride) + width]
pred = normalize_pred(crop_img_arr,crop_mask_arr)
crop_img_arr
full_prob[h * stride:(h * stride) + height,w * stride:(w * stride) + width] += pred[0]
full_sum[h * stride:(h * stride) + height,w * stride:(w * stride) + width] += 1
i = i + 1
out_img = full_prob / full_sum
return out_img
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--test_data', type=str, default='DRIVE', required=True, choices=['DRIVE','CHASE','STARE'])
parser.add_argument('--weight_name_global',type=str, help='path/to/global/weight/.h5 file', required=True)
parser.add_argument('--weight_name_local',type=str, help='path/to/local/weight/.h5 file', required=True)
parser.add_argument('--stride', type=int, default=3, help='For faster inference use stride 16/32, for better result use stride 3.')
args = parser.parse_args()
## Input dimensions
image_shape_fine = (128,128,3)
mask_shape_fine = (128,128,1)
label_shape_fine = (128,128,1)
image_shape_x_coarse = (64,64,128)
image_shape_coarse = (64,64,3)
mask_shape_coarse = (64,64,1)
label_shape_coarse = (64,64,1)
img_shape_g = (64,64,3)
ndf=64
ncf=128
nff=128
## Load models
K.clear_session()
opt = Adam()
g_local_model = fine_generator(x_coarse_shape=image_shape_x_coarse,input_shape=image_shape_fine,mask_shape=mask_shape_fine,nff=nff)
print("loc",args.weight_name_local)
g_local_model.load_weights(args.weight_name_local)
g_local_model.compile(loss='mse', optimizer=opt)
g_global_model = coarse_generator(img_shape=image_shape_coarse,mask_shape=mask_shape_coarse,ncf=ncf)
g_global_model.load_weights(args.weight_name_global)
g_global_model.compile(loss='mse',optimizer=opt)
## Find file numbers,paths or names
if args.test_data == 'DRIVE':
limit = 20
elif args.test_data == 'CHASE':
filenames = glob.glob("CHASE/test/images/*.jpg")
limit = len(filenames)
elif args.test_data == 'STARE':
arr = ["im0001","im0082","im0236","im0319"]
limit = 4
## Iterating for each image
#960,999
w,h =584,565
y_true = np.zeros((limit,w,h),dtype=np.float32)
y_pred = np.zeros((limit,w,h),dtype=np.float32)
y_pred_auc = np.zeros((limit,w,h),dtype=np.float32)
c = 0
for i in range(0,limit):
if args.test_data == 'DRIVE':
if i<9:
ind = str(0)
else:
ind = str("")
label_name = "/home/ubuntu/efs/imagesets/drive/test/1st_manual/"+ind+str(i+1)+"_manual1.gif"
print("file",label_name,i)
label = Image.open(label_name)
label_arr = np.asarray(label,dtype=np.float32)
img_name = "/home/ubuntu/efs/imagesets/drive/test/images/"+ind+str(i+1)+"_test.tif"
tif = TIFF.open(img_name)
img_arr = tif.read_image(tif)
img_arr = np.asarray(img_arr,dtype=np.float32)
mask_name = "/home/ubuntu/efs/imagesets/drive/test/mask/"+ind+str(i+1)+"_test_mask.gif"
mask = Image.open(mask_name)
mask_arr = np.asarray(mask,dtype=np.float32)
elif args.test_data == 'CHASE':
k = filenames[i].split('/')
k = k[-1].split('.')[0]
label_name = "CHASE/test/labels/"+k+"_1stHO.png"
label = Image.open(label_name)
label_arr = np.asarray(label,dtype=np.float32)
img_name = "CHASE/test/images/"+k+".jpg"
img = Image.open(img_name)
img_arr = np.asarray(img,dtype=np.float32)
mask_name = "CHASE/test/mask/"+k+"_mask.png"
mask = Image.open(mask_name)
mask_arr = np.asarray(mask,dtype=np.float32)
elif args.test_data == 'STARE':
label_name = "STARE/test/labels-ah/"+arr[i]+".ah.ppm"
label = Image.open(label_name)
label_arr = np.asarray(label,dtype=np.float32)
img_name = "STARE/test/stare-original-images/"+arr[i]+".ppm"
img = Image.open(img_name)
img_arr = np.asarray(img,dtype=np.float32)
mask_name = "STARE/test/mask/"+arr[i]+"_mask.png"
mask = Image.open(mask_name)
mask_arr = np.asarray(mask,dtype=np.float32)
## Get the output predictions as array
## Stride =3 (best result), Stride = 32 (faster result).
out_img = strided_crop(img_arr, mask_arr, mask_arr.shape[0], mask_arr.shape[1], 128, 128,args.stride)
print("cnt",np.count_nonzero(out_img),out_img.shape)
out_img[mask_arr==0] = 0
y_pred_auc[c,:,:] = out_img
out_img[out_img>=0.5] = 1
out_img[out_img<0.5] = 0
y_true[c,:,:] = label_arr
y_pred[c,:,:] = out_img
c = c +1
y_true[y_true==255]=1
y_true = y_true.flatten()
y_pred = y_pred.flatten()
y_pred_auc = y_pred_auc.flatten()
print("y_true",y_true.shape,np.unique(y_true),"y_pred",y_pred.shape,np.unique(y_pred))
confusion = confusion_matrix(y_true, y_pred)
print('confusion',confusion)
tn, fp, fn, tp = confusion.ravel()
metric_cal = time.time()
if float(np.sum(confusion)) != 0:
accuracy = float(confusion[0, 0] + confusion[1, 1]) / float(np.sum(confusion))
print("Global Accuracy: " + str(accuracy))
specificity = 0
if float(confusion[0, 0] + confusion[0, 1]) != 0:
specificity = tn / (tn + fp)
print("Specificity: " + str(specificity))
sensitivity = 0
if float(confusion[1, 1] + confusion[1, 0]) != 0:
sensitivity = tp / (tp + fn)
print("Sensitivity: " + str(sensitivity))
precision = 0
if float(confusion[1, 1] + confusion[0, 1]) != 0:
precision = tp / (tp + fp)
print("Precision: " + str(precision))
F1_score = 2*tp/(2*tp+fn+fp)
print("F1 score (F-measure): " + str(F1_score))
AUC_ROC = roc_auc_score(y_true, y_pred_auc)
print("AUC_ROC: " + str(AUC_ROC))
ssim = ssim(y_true, y_pred, data_range=y_true.max()-y_true.min())
print("SSIM: " + str(ssim))
meanIOU = jaccard_score(y_true,y_pred) #,pos_label=255.0) #normalize=True)
print("meanIOU: " + str(meanIOU))
from rvgan.
Related Issues (20)
- Inference time HOT 1
- Some questions about the paper and code HOT 9
- pretrained model of chase can't be loaded HOT 16
- Killed error HOT 1
- local_plot predictions always blank HOT 3
- How to determine the best model HOT 4
- need help! HOT 2
- Hello! HOT 5
- Hi~ HOT 1
- Having nan value for all the losses HOT 1
- length error! HOT 3
- libtiff error
- the generation of the final output image HOT 2
- When we run train.py, I find the loss is 'nan' at epchs 2. Do you have this problem. I want to know why and how to solve it. HOT 1
- Question HOT 12
- whether we need to pay attention to some details during training? HOT 12
- Question about loss decline HOT 2
- How should we train the model using tf 2.6.0? HOT 4
- Warning while training HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from rvgan.