Question about Figure 9

cambrian-mllm / cambrian

Cambrian-1 is a family of multimodal LLMs with a vision-centric design.

https://cambrian-mllm.github.io/

Apache License 2.0

1.4k stars 88 forks source link

Question about Figure 9 #11

Closed digbangbang closed 2 days ago

digbangbang commented 6 days ago

Why the Filtered DVQA has 1550K data more than the DVQA 775K data, the latter I suppose is unfiltered?

Also the situation on the CLEVR(Filtered CLEVR 350K same as CLEVR 350K)

tsb0601 commented 2 days ago

Hi, in Figure 9 and Cambrian-7M , we filtered out 1550K DVQA data and kept 775K DVQA data. The same applies to CLEVR. We used randomly sampling when we do filtering