video data's amount mismatch with the ola paper mentioned
#2
by
Kyousan
- opened
Hello, Thanks for the open-source training data!
In paper, I found that in stage 2 you mentioned using the 1.9M video data, but in the dataset it has 0.92M video data with 5.77M conversation pieces in video-data.json
and in stage 3 you mentioned using the 83k video data(filtered llava and finevideo) but in the dataset it has llava-123k and finevideo 175k
so is there something I misunderstood or some procedure needs to be done to process the data again?
Kyousan
changed discussion status to
closed