search

lucazanella / lavad

Official implementation of "Harnessing Large Language Models for Training-free Video Anomaly Detection", CVPR 2024
https://lucazanella.github.io/lavad/
48 stars 2 forks source link

Dataset development #4

Closed Libyte closed 3 months ago

Libyte commented 3 months ago

Thank you for your excellent work that has benefited me a lot, but I downloaded the original dataset annotation information and the annotation information given on google Driver during the reproduction process, and I would like to ask what should be done.

lucazanella commented 3 months ago

Hi, thank you!

I see you closed the issue, did you solve the problem?

After downloading the videos, you'll need to convert them to frames. You can use this script as a reference. For each video, the annotation file (e.g., test.txt) should include the relative path of the video (without extension) to the root_path directory, the start frame index, end frame index, and label index. The label index is not used, so you can set it to a placeholder or remove this entry from VideoRecord.

Let me know if this helps!