search

lilygeorgescu / AED-SSMTL

Anomaly Detection in Video via Self-Supervised and Multi-Task Learning
Other
56 stars 1 forks source link

Questions about frame-level detection #7

Closed hongsi96 closed 2 years ago

hongsi96 commented 2 years ago

Hi,

I have a question about the frame-level detection experiment.

In the case of object-level detection, the cropped image extracted with the YOLOv3 detector is resized to a 64 × 64 pixel image. And this is clear. But, In the case of frame-level detection, I wonder how to pre-process the original video (image). Are you resizing in the same way (64 × 64 pixel image)? Or is it some other preprocessing process?

It would be helpful to give the necessary information to reproduce the frame-level detection experiment.

Thank you, Sincerely yours,

Seokil

lilygeorgescu commented 2 years ago

Hi Seokil,

Apologize for my late reply. In the case of frame-level detection, we resized the images to 256x256. My co-author @antoniobarbalau can give you more details about this experiment.

Best, Lili

hongsi96 commented 2 years ago

Thank you for reply. Actually, I tried to resize it to 256x256, but the performance listed in the paper did not come out.

If there is additional information needed for frame-level detection experiment reproduction, can you share it?

@antoniobarbalau I would be very grateful if you could share any elements that could be a clue.