|
--- |
|
inference: false |
|
--- |
|
|
|
|
|
<br> |
|
<br> |
|
|
|
# ViP-LLaVA Model Card |
|
|
|
## Model details |
|
|
|
**Model type:** |
|
ViP-LLaVA is an open-source chatbot trained by fine-tuning LLaMA/Vicuna on both image level instruction data and region-level instruction data annotated with visual prompts. |
|
It is an auto-regressive language model, based on the transformer architecture. |
|
|
|
**Model date:** |
|
ViP-LLaVA-7B was trained in November 2023. [Paper](https://arxiv.org/abs/2312.00784) |
|
|
|
**Paper or resources for more information:** |
|
https://vip-llava.github.io/ |
|
|
|
## License |
|
Llama 2 is licensed under the LLAMA 2 Community License, |
|
Copyright (c) Meta Platforms, Inc. All Rights Reserved. |
|
|
|
**Where to send questions or comments about the model:** |
|
https://github.com/mu-cai/ViP-LLaVA/issues |
|
|
|
## Intended use |
|
**Primary intended uses:** |
|
The primary use of ViP-LLaVA is research on large multimodal models and chatbots. |
|
|
|
**Primary intended users:** |
|
The primary intended users of the model are researchers and hobbyists in computer vision, natural language processing, machine learning, and artificial intelligence. |
|
|
|
## Training dataset |
|
- 558K filtered image-text pairs from LAION/CC/SBU, captioned by BLIP. |
|
- 665K image level instruction data from LLaVA-1.5. |
|
- 520K image-text pairs marked with visual prompts. |
|
- 13K region-level instruction data generated from GPT-4V. |
|
|
|
## Evaluation dataset |
|
ViP-LLaVA achieves state-of-the-art performance in 4 academic region-level benchmarks and our newly proposed RegionBench. |
|
|
|
|
