sucv
commented
3 months ago Normal. Preprocessing includes many steps, the most time consuming step is transcribing.
To accelerate the proces, you may not need to generate all features. For visual, leave it as is. For audio, only need vggish and log Mel. for text, may consider discard it. Then it may need about 1day.
In my attempt, the Bert features actually slightly degrade the result. It is good for writing a fancy paper but may not good for achieving the best result, at least not in my method.
On Wed, Mar 13, 2024 at 18:30 Hanyu Cao @.***> wrote:
Hello, when I ran the code for data preprocessing, it took more than a week to complete. Is this normal? Additionally, I checked the GPU utilization in the background and found that it was almost not being used. However, torch.cuda.is_available() returns True, indicating that it's available.
— Reply to this email directly, view it on GitHub https://github.com/sucv/ABAW3/issues/9, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHBMSEDWJNAYPFLBGPNH4RDYYATFNAVCNFSM6AAAAABET4CJIOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGE4DGNRUGU3DONY . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Hello, when I ran the code for data preprocessing, it took more than a week to complete. Is this normal? Additionally, I checked the GPU utilization in the background and found that it was almost not being used. However, torch.cuda.is_available() returns True, indicating that it's available.