Need your help.

[Cola-any]

Hello, thank you for your excellent work MixGen and code sharing. I have two questions that I would like to call for your help:

1. In addition to being used in the pre-training process, is MixGen also used in downstream tasks, such as fine-tuning on the COCO data set for image-text retrieval task.
2. Considering that the augmentation operation is performed in mini-batch and there involves text concat, will this have an impact on the contrastive learning pre-training objective (as in common, other samples in mini-batch are regarded as negative samples).