search

MengLcool / DeepStack-VL

[NeurIPS-24] This is the official implementation of the paper "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs".
Apache License 2.0
32 stars 1 forks source link

Training data #1

Open John-Ge opened 5 months ago

John-Ge commented 5 months ago

hello, very nice work! Would you like to release your training data?

MengLcool commented 5 months ago

Thank you for your attention to our work. We will release the training data soon. Specifically, we used the LLaVA-1.5 training data for our experiments on DeepStack-L/V and mixed publicly available data for DeepStack-HD.