how did you run caption?

Question

how did you run caption?

Closed this issue 5 months ago · 5 comments

hello author，I ran 01_caption.sh on A6000 and found that it takes a long time (possibly several days) for XD_violence dataset to generate captions. How did you handle it? Is it useful for multi GPUs running?

Answer 1 · 2024-07-03T14:56:16.000Z

Hi, unfortunately, I also experienced long processing times to get output from BLIP-2. To address this, I sped up the process by distributing the videos across multiple compute nodes, as I had access to a machine with many nodes. What --batch_size are you using?

Answer 2 · 2024-07-03T15:12:33.000Z

batch_size I used 1, now I split the data set into multiple folders and multi-process running

Answer 3 · 2024-07-03T16:40:54.000Z

I also suggest increasing the batch size (using the --batch-size parameter) to maximize GPU utilization.

Answer 4 · 2024-07-03T16:44:53.000Z

Ok, I'll have a try. Thank you so much for taking the time to answer my questions.

Answer 5 · 2024-07-04T08:01:41.000Z

Thanks to you for your interest in our work :)