search

invictus717 / MetaTransformer

Meta-Transformer for Unified Multimodal Learning
https://arxiv.org/abs/2307.10802
Apache License 2.0
1.52k stars 114 forks source link

Is the model that each task will have a corresponding downstream HEAD MLP? #32

Closed moonriver0922 closed 1 year ago

moonriver0922 commented 1 year ago

Great work, thanks. I have a question that Is the model that each task will have a corresponding downstream HEAD MLP?

invictus717 commented 1 year ago

Thank you very much for your acknowledgment! For downstream tasks, Meta-Transformer requires relevant task-specific heads to adapt output formats. For example, it requires convolutional layers as the HEAD for image segmentation tasks and MLP layers for classification and other forecasting tasks.

moonriver0922 commented 1 year ago

Got it. Thanks!