def eval_net(net, loader, device, n_val):
"""Evaluation without the densecrf with the dice coefficient"""
print('eval in process')
net.eval()
tot = 0
accuracy = 0
dice_panck = 0
dice_nuclei = 0
dice_lcell = 0
dice_avr = 0
for batch in loader:
imgs = batch['image']
true_masks = batch['mask']
imgs = imgs.to(device=device, dtype=torch.float32)
true_masks = true_masks.to(device=device, dtype=torch.long)
true_masks = true_masks.squeeze(1)
# mask_pred, aux = net(imgs) #for deeplabv3 and deeplabv3+
mask_pred = net(imgs)
for true_mask, pred in zip(true_masks, mask_pred):
if net.n_classes > 1:
tot += F.cross_entropy(pred.unsqueeze(dim=0),
true_mask.unsqueeze(dim=0)).item()
else:
tot += dice_coeff(pred, true_mask.squeeze(dim=1)).item()
if net.n_classes > 1:
probs = F.softmax(pred, dim=1)
else:
probs = torch.sigmoid(pred)
probs = probs.cpu().detach().numpy()
true_mask1 = true_mask.cpu().numpy()
probs = np.argmax(probs, axis=0)
#probs = nuclei_process(probs)
accuracy += accuracy_score(probs, true_mask1)
dice_avr += diceCoeff_avr(probs, true_mask1)
dice_panck += diceCoeff_panck(probs, true_mask1)
dice_nuclei += diceCoeff_nuclei(probs, true_mask1)
dice_lcell += diceCoeff_lcell(probs, true_mask1)
return tot / n_val, accuracy / n_val, dice_avr / n_val, dice_panck / n_val, dice_nuclei / n_val, dice_lcell / n_val
#for ph2
fname = fname.split('_')[0]
#print(fname)
#save the images
x1 = plt.imsave('./model/pred_threshold/' + fname + '_thrPrediction.png', s)
y = Image.open(x_sort_testL[i])
s2 = data_transform32(y)
s3 = np.array(s2)
# s2 =threshold_predictions_v(s2)
#save the Images
y1 = plt.imsave('./model/label_threshold/' + fname + '_thrLabel.png', s3)
total = dice_coeff(s, s3)
acc, sen, spe, dice, jacc = accuracy_score(s, s3)
names = x_sort_testP[i].split('/')
logInfoHead = names[-1]+'(dice1): '+str(total)
logInfoBody = 'accuracy:%.5f, sensitivity:%.5f, specificity:%.5f, dice:%.5f, jaccard:%.5f'%(acc, sen, spe, dice, jacc)
f.write(logInfoHead+"\n")
f.write(logInfoBody+"\n")
print(logInfoHead)
print(logInfoBody)
t_acc += acc
t_sen += sen
t_spe += spe
t_dice += dice
t_jacc += jacc
if total <= 0.65:
x_count += 1
f2.write("name:%s, dice:%.5f"%(names[-1], total)+"\n")
dice_lcell = 0
for filename in os.listdir(in_files):
count = count + 1
img = Image.open(in_files + "/" + filename)
img_array = np.array(img)
img = img_array
mask = predict_img(net=net,
full_img=img,
scale_factor=1.0,
use_dense_crf=False,
device=device)
mask = nuclei_process(mask)
true_mask = Image.open(true_masks + "/" + filename[:-4] + ".bmp")
true_mask = np.array(true_mask)
accuracy += accuracy_score(mask, true_mask)
precision += precision_score(mask, true_mask)
recall += recall_score(mask, true_mask)
f1 += f1_score(mask, true_mask)
iou += IOU(mask, true_mask)
diceCoeff += diceCoeff_avr(mask, true_mask)
dice_panck += diceCoeff_panck(mask, true_mask)
dice_nuclei += diceCoeff_nuclei(mask, true_mask)
dice_lcell += diceCoeff_lcell(mask, true_mask)
mask_viz = mask_to_image(mask)
#imageio.imwrite(output + "/" + filename, mask_viz)
mask = Image.fromarray((mask).astype(np.uint8))
imageio.imwrite(output + "/" + filename[:-4] + '.bmp', mask)
print("num of samples%.0f" % count)
print("accuracy = %.3f" % (accuracy / count))
# digits.target_names
# some_digit = X[666] # 随机取一个样本
# some_digmit_image = some_digit.reshape(8, 8)
# plt.imshow(some_digmit_image, cmap = matplotlib.cm.binary)
# plt.show()
# y[666]
# knn进行分类
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.2,
random_state=666)
knn_clf = KNeighborsClassifier(n_neighbors=5)
knn_clf.fit(X_train, y_train)
y_predict = knn_clf.predict(X_test)
acc = accuracy_score(y_test, y_predict)
print(acc)
# 直接调用检验预测的方法
acc = knn_clf.score(X_test, y_test)
# 指定最佳值的分数，初始化为0.0；设置最佳值k，初始值为-1
best_score = 0.0
best_k = -1
for k in range(1, 11): # 暂且设定到1～11的范围内
knn_clf = KNeighborsClassifier(n_neighbors=k)
knn_clf.fit(X_train, y_train)
score = knn_clf.score(X_test, y_test)
print(score)
if score > best_score:
best_k = k
best_score = score
im_tb = Image.open('/home/malav/Desktop/Pytorch_Computer/DATA/test_new_3C_I_ori_same/0131_0009.png')
im_label = Image.open('/home/malav/Desktop/Pytorch_Computer/DATA/test_new_3C_L_ori_same/0131_0009.png')
s_tb = data_transform(im_tb)
s_label = data_transform(im_label)
pred_tb = model_test(s_tb.unsqueeze(0).to(device)).cpu()
pred_tb = F.sigmoid(pred_tb)
pred_tb = pred_tb.detach().numpy()
#pred_tb = threshold_predictions_v(pred_tb)
x1 = plt.imsave(
'./model/pred/img_iteration_' + str(n_iter) + '_epoch_'
+ str(i) + '.png', pred_tb[0][0])
accuracy = accuracy_score(pred_tb[0][0], s_label)
#######################################################
#To write in Tensorboard
#######################################################
train_loss = train_loss / len(train_idx)
valid_loss = valid_loss / len(valid_idx)
if (i+1) % 1 == 0:
print('Epoch: {}/{} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}'.format(i + 1, epoch, train_loss,
valid_loss))
writer1.add_scalar('Train Loss', train_loss, n_iter)
writer1.add_scalar('Validation Loss', valid_loss, n_iter)
#writer1.add_image('Pred', pred_tb[0]) #try to get output of shape 3
def score(self, X_test, y_test):
"""根据测试数据集,X_test和y_test确定当前模型准确度"""
y_predict = self.predict(X_test)
return accuracy_score(y_test, y_predict)