[Question] About Data Augmentation

[ia-gu]

Question

I would like to introduce data augmentation to images (texts too in the near future), but I cannot find transforms.
Where can I introduce data augmentation?
Thanks in advance.

[mu-cai]

Hi,

I guess you are referring that how I edit the images. Here is the code https://github.com/WisconsinAIVision/ViP-LLaVA/blob/main/llava/visual_prompt_generator.py

[ia-gu]

Sorry, I meant to refer to common data augmentation methods (e.g., HorizontalFlip(), Normalize()).

I supposed to say that https://github.com/WisconsinAIVision/ViP-LLaVA/blob/main/llava/visual_prompt_generator.py does not include them.

[mu-cai]

There is no data augmentation.

For each image, there are only two preprocess steps:

• Padding: https://github.com/WisconsinAIVision/ViP-LLaVA/blob/main/llava/train/train.py#L900
• Normalizetion: https://github.com/WisconsinAIVision/ViP-LLaVA/blob/main/llava/train/train.py#L901, which is hidden in the transformers clip encoder.

[ia-gu]

I didn't know that.
Thanks!!!