Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

发现程序中一个错误 #1

Open
zezisme opened this issue Dec 19, 2019 · 10 comments
Open

发现程序中一个错误 #1

zezisme opened this issue Dec 19, 2019 · 10 comments

Comments

@zezisme
Copy link

zezisme commented Dec 19, 2019

在tiny-yolov1.py脚本中,第150行代码:if iou_score >0.2:判断iou值大小这里应该是小于号,即:if iou_score < 0.2:

@jcui65
Copy link

jcui65 commented Jan 2, 2020

我刚刚把代码下载下来,正在进行各种尝试。我还没很理解这个代码。但是试了大于0.2和小于0.2之后,我不觉得原来的代码是个错误。

@jcui65
Copy link

jcui65 commented Jan 2, 2020

我觉得原来的代码是对的,如果IOU>0.2,则说明该部分(该物体块)与当前的最大的物体块的重合度够高(大于0.2),于是,这个物体块不属于一个单独的新物体,故而将其filter_mask置零,也就是消除了一个多余的,无用的物体。

@zezisme
Copy link
Author

zezisme commented Jan 7, 2020

我觉得原来的代码是对的,如果IOU>0.2,则说明该部分(该物体块)与当前的最大的物体块的重合度够高(大于0.2),于是,这个物体块不属于一个单独的新物体,故而将其filter_mask置零,也就是消除了一个多余的,无用的物体。

是的,你是对的,是我当时没理解对,不过这个github项目貌似有点问题,不知道你跑出来没有,我不管是用voc数据集还是自己的,训练很多轮后,预测结果都特别差,而且预测的结果貌似都一样,跟训练不训练没有区别,都是那几个框,完全跟没读过图像信息一样

@jcui65
Copy link

jcui65 commented Jan 7, 2020

我跑了30个epoch,感觉还可以。当然,我可能就挑了几张图片。你训练了多少个epoch呢?你测试了多少图片呢?

@BryceYuZh
Copy link

我在voc数据集上训练了30epoch之后 他似乎并不能训练出想要的结果

@AlucardNosferatu
Copy link

我在voc数据集上训练了30epoch之后 他似乎并不能训练出想要的结果

我也是,一言不合就person,trust score特别低

@AlucardNosferatu
Copy link

我在VOC2007全体数据跑了120个epoch,测试时效果还是很差
YOLOv1-000002
YOLOv1-000003
YOLOv1-000004

@AlucardNosferatu
Copy link

训练200轮以上,似乎有所改观:
YOLOv1-000002
YOLOv1-000003
YOLOv1-000004
目前loss和val_loss仍在下降,等收敛后再看看

@AlucardNosferatu
Copy link

确认模型应该是没问题 ,尝试换VGG16做backbone也能正常预测,但是太容易过拟合,加了BN也没用,可能还得换回原配的Backbone试试

@georgenardes
Copy link

I have trained for more than 50 epochs with one single image and the output box doesn't get closer to the ground truth.
I think that it should overfit the data, but it is converging to a wrong value. I put a "K.print_tensor(y_pred[0, 3, 3])" in the loss function to see what is happening and the same values are predicted in all tries. The index [0,3,3] is the grid in which the object center is contained.
Has anyone got trough this like me?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants