yongliu20/qdmn

[ECCV2022] Learning Quality-aware Dynamic Memory for Video Object Segmentation

qdmn's Issues

您好,当我将QAM加入到STCN当中时,且输入到Score模块的特征是fuse后的,但是训练得到的score值到后面都是0了(Score模块里面全连接层的激活函数一个是relu,一个是sigmoid)我是1GPU训练,batch size为16,学习率为1e-5

I use QAM for training, the number of iterations is 75K, but the loss only drops to about 0.5, is this normal?And how much does the final loss value of the original code drop?

您好，注意到您的代码变了，请问是之前的代码有问题吗？

Is the input dimension of Score 256 when QAM is used in STCN?

train.py启动问题

您好，请问单GPU情况下，您给出的train.py启动方式中，参数值需要怎么改？

Questions about pre-training

Hello, if you only use static_ Can images (stage-0) and daivs-youtube (stage-2) be trained?
How much worse is the accuracy of the model than that of the model containing BL30K?

The code has been reproduced, but the results are inconsistent with the provided model.

Why does the network I reproduced from the code have a score of 0 when tested, while testing with the provided pth file results in a non-zero score?

About Score class

关于预训练的问题

您好，我想问一下，stage-0 是静态图片，这个静态图片的数据集是哪一个呢？stage-1是BL30k，这个数据集的处理和davis是一致的嘛？

About Score class

Hello! I encountered some issues while training the code and would like to seek your advice. When I was training the code, the score I got was very low (7.9036e-5). I would like to ask if you have also encountered such a situation. Looking forward to your reply, thank you very much。

您好，请问论文中为什么不使用L2 distance而是dot product呢，是L2 distance效果比dot product差吗

请问预训练阶段是不需要调整学习率的吗？代码中预训练时的steps比它的迭代次数还要大

youtube2018

您好，请问如何对youtube2018的结果进行评估，您所提供的链接失效，无法进行评估？

为什么eval代码要用到label?

Do you train the whole network when you apply QAM in STCN?

Do you train the whole network when you apply QAM in STCN? or you just load the weights of the initial model (trained without QAM) and load the parameters of QAM trained in QDMN, like what you do in Table 4.

checkpoint file

Great job! Could you provide pretrained checkpoint file for test. Thank you!

May ask the result when you use the global average pooling layers in the code of QAM module?

May I ask how much the effect will probably decrease？

When the QAM is used in STCN,is the iterations the same as STCN?

Adding QAM only in STM's inference stage

您好，感谢您的这项出色的工作！
我现在在复现论文中所说的将QAM仅用于STM的测试阶段，但是有如下困惑，希望您能不吝指教：
1.将QAM用于STM测试阶段时，记忆池更新机制是否也要使用？
2.是对每一视频帧都用QAM计算其质量得分＞0.8即可，还是像您代码中那样，至少间隔5帧再判断？