search

mira-space / MiraData

Official repo for paper "MiraData: A Large-Scale Video Dataset with Long Durations and Structured Captions"
https://mira-space.github.io/
GNU General Public License v3.0
349 stars 9 forks source link

Training data for MiraDiT #15

Open rhjdzuDL opened 3 weeks ago

rhjdzuDL commented 3 weeks ago

Great work! However, I noticed that there are some missing details regarding the model training in the paper (or perhaps I overlooked them; were they disclosed elsewhere?). I would greatly appreciate it if you could clarify them:

  1. Which version of the data was used to train the MiraDiT model (the results in Table 3)? Was it the 330K, 93K, 42K, or 9K version?
  2. Which version of the captions was used during training? Was it the short, dense, or structural captions?
  3. What is the total amount of training data?

Thank you in advance for your response!