Closed ElegantLin closed 10 months ago
hi our current implementation use different args to load different groups of datasets.
For video datasets, you should load it with mimicit_vt_path
(others also follow the vt
tag).
For general image-text, you should load it with mimicit_path
(no specific tag).
We will have a next release maybe late this month or early next month. There will be a big update in training pipeline, and it would better support mixture training. If you do have interests using our codebase, you could send me an email and we could discuss more on that.
Thanks for your quick reply. My task is only about image-text at this time so I should only use mimicit_path
. But I understand the current version cannot support mixture training and I can only pass one json
file here. Is that correct?
Thanks!
HI authors,
Thanks for your great work. It seems that MIMIC-IT is a dataset consisting of several sub-datasets. Otter benefits from all these datasets and can preform well in several down-stream VL tasks. However, in the example,
mimicit_path
,images_path
,train_config_path
can only accept one json file, respectively. I wonder whether it is possible to set several json files for these arguments. For example, can I set the following arguments?The tag of this issue should be
train
.Thanks!
Before you open an issue, please check if a similar issue already exists or has been closed before.
When you open an issue, please be sure to include the following
dataset
(mimic-it download, usage, etc.),demo
(online demo),doc
(readme, wiki, paper, video etc.),evaluation
(evaluation result, performance of Otter etc.),model
(model configuration, components, etc.),train
(training configuration, process, code, etc.)Thank you for your contributions!