okankop / driver-anomaly-detection

PyTorch Implementation of "Driver Anomaly Detection: A Dataset and Contrastive Learning Approach", codes and pretrained models.

Hi there, I'm Okan okankop.github.io 👋

---

I'm a PhD Student at Technical University of Munich! I am supervised by Prof. Gerhard Rigoll. Before that, I received my Master's degree in Electrical Engineering from Technical University of Munich and my Bachelor's degree from Boğaziçi University in 2017 and 2014, respectively.

My research interests lie in computer vision, deep learning, and their applications. Specifically on tasks related to video understanding, gesture recognition and action localization.

---

---

driver-anomaly-detection's Issues

a trained pth file

Hi Okan, can you send me a trained pth file, because the dataset is too big, I am a student and I can't train it on my own computer...

How could you determine the hyperparameter value on the test set?

Hi @okankop ,

I think you might have made an error in the code for validation/testing. It seems that the value of threshold is determined by the results on validation/testing. If you split the original test set into a validation set and a new test set, then it's fine to tune this hyperparameter on the validation set. But the problem is, it seems that you didn't do it...

Many thanks,
Yiming

Data setting

Hi Okan,
Interesting work!

I try to use your model in the application. But my RGBD camera is realsense, which depth picture seems different from yours. So, would you please tell me the data setting of your rgbd-camera, such as the grayscale corresponding to the true depth value.

Fail to decompress the data

I have failed to decompress the rar file. How do I decompress the RAR file?

Dataset with RGB frames

Hi Okan,

Interesting work!
I am quite curious about the reason why you only collect the depth and infrared frames.
I believe that RGB frame is more widely used in general action recognition task. Could you please illustarte more?

Action label for test set

Hello,

Wonderful work! I am wondering if you can also provide the action label for the test set as well. So far there is only anomaly and normal.

Thank you

More RAM is consumed as training epoch increases

I have repeated the experiment for many times and eventually at the 14th epoch, the process is killed, mainly due to my CPU RAM is exhausted. My current configuration is 8 core CPUs, 30GB CPU RAM, P5000 16GB GPU.

I have tried to reduce memory bank size, but it's still the same.

How do you split the training set and test set

This is a great project. I learn a lot from it. However, I have a little question. How do you split the training set and the test set? I have seen the DAD class and DAD_Test class for loading the dataset, but it seems to load the same data for training and testing. Can you give me a hand about this?

Dataset

Dear authors, your work is very interesting. Could you please provide with your dataset? How did you prepare and label data?

Dataset decompression problem

Hello okankop，JiapengZh，eniolasonowo,
Thanks for your best work!
I have some questions about the dataset decompression. The trouble is that when I decompressed it, I found that some data could not be decompressed or decompressed incorrectly. Have previous researchers encountered similar problems? Is there a solution to this?
Thank you in advance for any help you might be able to provide.
Sincerely,
Sally Guo

dataset structure

hello @okankop @JiapengZh @eniolasonowo
Thanks for your best work!
I have some questions about dataset structure, can you share the dataset structure with me ? I want to know the files contained in each video folder

Thanks a lot!