I found that they are almost the same...

CPD:

NCD:

作者您好，非常感谢您的优秀成果和开源精神！
如果我的问题看起来不是那么的“聪明”，还请谅解。
我想知道论文中的Coarse Map 是怎么生成的，它的代码是什么？我自己尝试了一下（图二）发现它与论文中给出的Coarse Map（图一）完全不同。
最后，再次感谢您。

how is my database structured?

你好！预测结果可以进行mask AP 的评测吗？

Hi, authors, thank you very much for your great work. I downloaded the training/validation and testing data from your provided Google drive links. But I find the numbers do not match the figures in your PAMI paper. The number of training/validation images is 4040 rather than 6000+1250 in your paper. The number of testing images for COD10K is 2026 rather than 4000 in your paper. What is the problem? Could you please help to solve it? Thank you so much!

Dear Authors,

thanks for sharing the code.

Please let me know, if this line is correct
gt = randomPeper(gt)

should this not be applied to the image instead?

Thanks

Alexander

Hello, I have noticed that there are already instance-level image annotations in COD10K. Can SINet support instance-level segmentation now? If not, I want to try to improve it to the instance level. Please provide your suggestions. Thank you.

您好，我在使用计算指标的代码时，经常出现多个函数找不到的问题，如 未定义与 'double' 类型的输入参数相对应的函数 'bwdist'。 想请教下是matlab版本问题还是其他原因？谢谢！

Thanks for your sharing. I find that in your Mytrain_val.py, a validation dataset is utilized. I am not sure whether you use the CAMO for validation since the default val_root is './Dataset/TestDataset/CAMO/' . Besides, the training/validation dataset link only contains the training set. Could you please tell me the training details?

您好，
当评估模型时出现NaN错误，这种情况怎么解决呢？

作者您好，我看论文里的TEM模块图，对多个支路拼接后，使用了1*1conv。但是在代码中，对多支路concat后，conv_cat使用了3*3Conv，请问是以代码为准么？

Do you provide the result map corresponding to the test set? I don't seem to see it on github

我对你们的研究很感兴趣，文章写非常好，很详细。我想知道模型推理FPS是多少？

It's a great job！

Question1： What is the difference between RF and TEM? In my opinion, RF and TEM are the same。

SINetv1

SINetv2

Question2： I want to use the TEM module, which part of the network structure works best? (Is it OK to put the TEM module behind the backbone?) ）

This may be a simple question, but it is important to me, hope to get a response from you, thank you!

Hello, read a lot of your articles, I feel that you have a deep knowledge in camouflage biological detection, I want you a question: most of the current camouflage object detection is based on saliency detection as a method, and finally presented in a way like "segmentation", do you think that the detection of camouflage organisms in the form of bounding boxes can be?

Dear author, I would like to ask, in the process of paper re-emergence, according to the statement of the article (except for batch size replacement for 20) trained model of other indicators and the paper is very close, except E-measure difference is relatively large, is this in the acceptable range?
SINet_V2_Results：
CHAMELEON ( 76 images): S: 0.894, E: 0.934, F: 0.825, M: 0.030
CAMO ( 250 images): S: 0.817, E: 0.870, F: 0.737, M: 0.072
COD10K (2026 images): S: 0.813, E: 0.864, F: 0.676, M: 0.036