Could you please guide me in understanding how to TEST my own videos using this model. I'm new. I am unable to interpret how the data is labelled in the Test_Anomalyv2.txt file. I want to add my own videos in it to test but I have no idea how to. Any kind of help would be highly appreciated!

test_anomalyv2.txt

Is your evaluation strategy the same as the official?

I notice you average the AUCs of 140 testing abnormal videos. If I didn't remember wrong, the official calculated the whole videos at the same time.

Hai @seominseok0429

I tried to run code its works fine but i have an error on vis.py

Traceback (most recent call last):
  File "vis.py", line 10, in <module>
    from models.model import generate_model
ModuleNotFoundError: No module named 'models'

Thanks

You cannot calculate the AUC score in this way, as it leads to a higher overall AUC in the end.

Since you increment the AUC for every new addition of a video pair, the early increments are closer to 1 than 0, which inflates the final AUC. If anyone wishes to test this, use a small subset of the dataset and you will see very high AUC scores.

Thank you very much for your work. In order to get the same effect as you.I want to know how to process and train the original video data to get the segment features you share. Can you share the code?

The last 10 lines in train_normal.txt is duplicate, as below:

Normal_Videos_event/Normal_Videos016_x264.mp4 Normal_Videos_event/Normal_Videos017_x264.mp4 Normal_Videos_event/Normal_Videos020_x264.mp4 Normal_Videos_event/Normal_Videos825_x264.mp4 Normal_Videos_event/Normal_Videos826_x264.mp4 Normal_Videos_event/Normal_Videos827_x264.mp4 Normal_Videos_event/Normal_Videos856_x264.mp4 Normal_Videos_event/Normal_Videos857_x264.mp4 Normal_Videos_event/Normal_Videos858_x264.mp4 Normal_Videos_event/Normal_Videos859_x264.mp4

Hello! I am new to python and NN. I wanted to know for how many EPOCH should I train this model? Also, is there any limit to how low the LOSS can be? And should one stop training if the LOSS is not reducing? Any other information a newbie should have regarding the LOSS VALUE and the EPOCH?

Any type of comments would be highly appreciated! Cheers!

-Zain

Hello

def MIL(y_pred, batch_size, is_transformer=0):
    loss = torch.tensor(0.).cpu()
    loss_intra = torch.tensor(0.).cpu()
    sparsity = torch.tensor(0.).cpu()
    smooth = torch.tensor(0.).cpu()
    if is_transformer==0:
        y_pred = y_pred.view(batch_size, -1)
    else:
        y_pred = torch.sigmoid(y_pred)

    for i in range(batch_size):
        anomaly_index = torch.randperm(30).cpu()
        normal_index = torch.randperm(30).cpu()

Just change cuda to cpu it's working without Nvidia driver

Thank you

I'm sorry can you explain me how you generated ckpt.pth in the VIS directory? It's different from the one you can generate using the main with default settings

Thanks for your great work! You have submit the full code by the https://drive.google.com/file/d/1xYsBiCSmXjE0BwoiH_Bcm4AWaA1pESHF/view.However,when I run the vis.py, it's wrongModuleNotFoundError: No module named 'models.model'I think the model directory doesn't appear the model and can not impot generater.So,would you please help me? thx so much

Hai @seominseok0429

I am currently working on your code and it's amazing. The training part is working properly and visualize part I am facing an error while running vis.py

i got this error

if I am correct this error occurred due to image size. how to resolve this issue

PIL.UnidentifiedImageError: cannot identify image file '/NGC-DL-CONTAINER-LICENSE'

thanks

Hi, thanks for your amazing work!

I have been trying to plot the ROC curve for the performance evaluation, but I am confused about how the auc is calculated in the test.py. Could you kindly explain your method? For example, what does the number 140 mean in the equation auc/140 in the python file test.py?

`score_list3 = np.concatenate((score_list, score_list2), axis=0) gt_list3 = np.concatenate((gt_list, gt_list2), axis=0)

fpr, tpr, thresholds = metrics.roc_curve(gt_list3, score_list3, pos_label=1) auc += metrics.auc(fpr, tpr)

print('auc = ', auc/140)`

Additionally, would it be possible to explain some of your idea for how to show the ROC curve?