Hello, thank you so much for sharing such great code. I have a question. Your code is running at about 50 FPS, while the speed given in the original paper is 25 FPS. What do you think caused this? Is PyTorch faster? The original paper used TITAN, and I used 1080Ti.

Hello, I have seen that you have provided two kinds of flownet options. I would like to ask which performance is better. I tested it. It seems that Lite flownet and flownet2sd have similar performance. Is that right?

Another question, I read your commits history. You tried SGD optimizer, and then you changed back to Adam. Is Adam better in your experiment?

Hi! Thanks for your work! And I have to questions.
First, what I want to ask is that if I want to test in other dataset, such as Mall, what should I do?
And second, if I want to taste to make a new dataset, how should I label it?
Looking forward to your reply, thank you again.

hello, could you give a link of pretrained model of litenet and flownet?

Isn't the evaluation calculation wrong?
Considering Ped2, why are the scores normalized between 0 and 1 for each of the 12 video clips individually? This isn't correct as the camera view is the same in all of the clips.

hi,before training,you said"Download the trained model and put them under the 'weights' folder"
but i didn't find the 'weights' folder,could you tell me where the 'weights'? folder is?
thanks!

Hi！Do you use the pre-training weights of the lite-flownet author ? Also, which version of lite-flownet are you using？thank you very
much！

你好，请问如何自定义自己想要可视化某一张图片？（把自己想要的某张图片的target_frame），另外在可视化界面中，怎么调整target_frame上面的黄色横线的尺度单位？
很感谢你的工作！期待回复！

网上找的shanghaitech数据集都没有.mat标签文件，请问哪里可以找到

I run train.py a night. But nothing was showed on the terminal.how long do you train?

I want to know how to get AUC value? Thanks.

When I run python train.py --dataset=avenue,The following problems arise

Shouldn't the channel information be part of frames when you are reading frames with 3 channels?

That is (n, 12, 256, 256) should be (n, 12, 3, 256, 256). Or is it that your code will only work with inputs of 1 channel?

EDIT: Or do we just multiply the channel information with clip length?
EDIT2: Got it!! :)

hello,have you run experiments on other datasets (like ped1 and shanghaitech)? Is there a weights file for these two datasets?

Hello，
After I read your code，I found that the unet.py does not implement the Concatenate operation,as the picture shows.
Am I right?
Thanks for your reply!

What changes have you made in the optical stream and the generator network? Can you describe it in detail? I hope to get your reply. Thank you.

The work is very niubility, and I appreciate it. But when i exec the "train.py" for about 100 epoch, the terminal display "segmentation fault (core dumped)". I try to solve this problem, but i failed. Anyone can help me? thank you very very very very muchhhhh!

我在训练时卡在dataloader循环之前。我注意到你在另一个工作 pre_ano_detection那边也有相同的问题，程序卡住下不去，你的这份代码我也遇到了同样的问题，请问你知道原因吗

Thanks for your project, i have run the train and evaluate , but how to output the abnormal frame?

Hello, how can I adjust the learning rate if I change the network structure? I am a beginner in deep learning. If you could give me some experience, I would be very grateful.

Hello，Could you tell me how to compute Equal Error Rate?
By the way,do you have any idea to improve your prediction model?
Have you tried to add a reconstruction model after the prediction model?

Another question：For example，after I trained the Avenue/training datasets,how to do test the Avenue/testing datasets?

Hello,Thanks for your work!
I have a question,when I running the "train.py" in pycharm,it no problem.
However,when I used Linux terminal run "python train.py --datasets=avenue",it said "ModuleNotFoundError: No module named 'cupy' ".But my packages including the 'cupy'.Thus,I don't know where is the problem.
I'm looking forward for your response.Thanks!

First of all, thank you for such a clean implementation and comments at different places to understand what's going on. Kudos and so much to learn. Also, cheers for still addressing issues! :)

So, in your Dataset.py file are you sampling multiple sequences or just in order the frames appear?
For example, if video frames are [01, 02, 03, 04, 05, 06, 07, 08, 09, 10] then are you doing (assuming input video length 5):

1. [01, 02, 03, 04, 05] and then[06, 07, 08, 09, 10]?
2. (or) Randomly sampling a sequence of 5 consecutive frames from the video (random start point).

If you doing the latter (point 2), are you sampling multiple times (in order to get more possible sequences from the same video)?

hi,when i used shanghaitech dataset for training, there has been no response since the program started
but this code works well on avenue and ped2

hi, when i trained this net, it showed"segmentation fault(core dumped)"
i want to know the specific version of your python,pytorch,cuda,thanks!