ask help - Githubissues

Yffy123456 commented 2 years ago

I would like to ask the author, when running step2to4_train_validate.py, the train [Validation] Acc: 0.9534, SE: 0.0000, SP: 0.0000, PC: 0.0000, Dice: 0.0000, IOU: 0.0000 Why do I get these results is 0?

WAMAWAMA commented 2 years ago

you can refer to the issue #8 and #14, and modify the evaluation code as follows


import numpy as np

def get_accuracy(SR,GT,threshold=0.5):
    SR = SR.view(-1)
    GT = GT.view(-1)

    SR = SR.data.cpu().numpy()
    GT = GT.data.cpu().numpy()

    SR = SR > threshold
    GT = GT == np.max(GT)
    corr = np.sum(SR==GT)

    acc = float(corr)/float(SR.shape[0])

    return acc

def get_sensitivity(SR,GT,threshold=0.5):
    SR = SR.view(-1)
    GT = GT.view(-1)

    SR = SR.data.cpu().numpy()
    GT = GT.data.cpu().numpy()

    SR = (SR > threshold).astype(np.float)
    GT = (GT == np.max(GT)).astype(np.float)

    # TP : True Positive
    # FN : False Negative
    TP = (((SR == 1.).astype(np.float) + (GT == 1.).astype(np.float)) == 2.).astype(np.float)
    FN = (((SR == 0.).astype(np.float) + (GT == 1.).astype(np.float)) == 2.).astype(np.float)

    SE = float(np.sum(TP)) / (float(np.sum(TP + FN)) + 1e-6)

    return SE

def get_specificity(SR,GT,threshold=0.5):
    SR = SR.view(-1)
    GT = GT.view(-1)

    SR = SR.data.cpu().numpy()
    GT = GT.data.cpu().numpy()

    SR = (SR > threshold).astype(np.float)
    GT = (GT == np.max(GT)).astype(np.float)

    # TN : True Negative
    # FP : False Positive
    TN = (((SR == 0.).astype(np.float) + (GT == 0.).astype(np.float)) == 2.).astype(np.float)
    FP = (((SR == 1.).astype(np.float) + (GT == 0.).astype(np.float)) == 2.).astype(np.float)

    SP = float(np.sum(TN)) / (float(np.sum(TN + FP)) + 1e-6)

    return SP

def get_DC(SR,GT,threshold=0.5):
    # DC : Dice Coefficient
    SR = SR.view(-1)
    GT = GT.view(-1)

    SR = SR.data.cpu().numpy()
    GT = GT.data.cpu().numpy()

    SR = (SR > threshold).astype(np.float)
    GT = (GT == np.max(GT)).astype(np.float)

    Inter = np.sum(((SR+GT)==2).astype(np.float))
    DC = float(2*Inter)/(float(np.sum(SR)+np.sum(GT)) + 1e-6)

    return DC

Yffy123456 commented 2 years ago

Hello, sorry to bother you, See your article 《Automatic Segmentation and Classification of Thyroid Nodules in Ultrasound Images with Convolutional》 on Springer Unfortunately, I have to pay a fee. I am still a student, so I can't pay a Neural network. So I was wondering if you could send me a copy. I'm sorry to bother you.

------------------ 原始邮件 ------------------ 发件人: "WAMAWAMA/TNSCUI2020-Seg-Rank1st" @.>; 发送时间: 2022年8月19日(星期五) 下午2:14 @.>; @.**@.>; 主题: Re: [WAMAWAMA/TNSCUI2020-Seg-Rank1st] ask help (Issue #35)

you can refer to the issue #8 and #14, and modify the evaluation code as follows import numpy as np def get_accuracy(SR,GT,threshold=0.5): SR = SR.view(-1) GT = GT.view(-1) SR = SR.data.cpu().numpy() GT = GT.data.cpu().numpy() SR = SR > threshold GT = GT == np.max(GT) corr = np.sum(SR==GT) acc = float(corr)/float(SR.shape[0]) return acc def get_sensitivity(SR,GT,threshold=0.5): SR = SR.view(-1) GT = GT.view(-1) SR = SR.data.cpu().numpy() GT = GT.data.cpu().numpy() SR = (SR > threshold).astype(np.float) GT = (GT == np.max(GT)).astype(np.float) # TP : True Positive # FN : False Negative TP = (((SR == 1.).astype(np.float) + (GT == 1.).astype(np.float)) == 2.).astype(np.float) FN = (((SR == 0.).astype(np.float) + (GT == 1.).astype(np.float)) == 2.).astype(np.float) SE = float(np.sum(TP)) / (float(np.sum(TP + FN)) + 1e-6) return SE def get_specificity(SR,GT,threshold=0.5): SR = SR.view(-1) GT = GT.view(-1) SR = SR.data.cpu().numpy() GT = GT.data.cpu().numpy() SR = (SR > threshold).astype(np.float) GT = (GT == np.max(GT)).astype(np.float) # TN : True Negative # FP : False Positive TN = (((SR == 0.).astype(np.float) + (GT == 0.).astype(np.float)) == 2.).astype(np.float) FP = (((SR == 1.).astype(np.float) + (GT == 0.).astype(np.float)) == 2.).astype(np.float) SP = float(np.sum(TN)) / (float(np.sum(TN + FP)) + 1e-6) return SP def get_DC(SR,GT,threshold=0.5): # DC : Dice Coefficient SR = SR.view(-1) GT = GT.view(-1) SR = SR.data.cpu().numpy() GT = GT.data.cpu().numpy() SR = (SR > threshold).astype(np.float) GT = (GT == np.max(GT)).astype(np.float) Inter = np.sum(((SR+GT)==2).astype(np.float)) DC = float(2*Inter)/(float(np.sum(SR)+np.sum(GT)) + 1e-6) return DC

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

WAMAWAMA commented 2 years ago

@Yffy123456 ok，just give me your Email😊

Yffy123456 commented 2 years ago

my @.*** very grateful.

------------------ 原始邮件 ------------------ 发件人: "WAMAWAMA/TNSCUI2020-Seg-Rank1st" @.>; 发送时间: 2022年8月21日(星期天) 中午11:59 收件人: @.>; 抄送: @.**@.>; 主题: Re: [WAMAWAMA/TNSCUI2020-Seg-Rank1st] ask help (Issue #35)

@Yffy123456ok，just give me your Email😊

— Reply to this email directly, view it on GitHub , or unsubscribe . You are receiving this because you were mentioned.Message ID: @.***>

WAMAWAMA commented 2 years ago

@Yffy123456 please give your email in CODE format like abc123@gmail.com, otherwise GitHub will mask it.

WAMAWAMA / TNSCUI2020-Seg-Rank1st

ask help #35