BLIVA_FlanT5 / README.md
gordonhubackup's picture
Merge branch 'main' of https://huggingface.co/mlpc-lab/LoViM_FlanT5
4f41567
|
raw
history blame
1.69 kB
metadata
license: bsd-3-clause
language:
  - en
pipeline_tag: visual-question-answering
library_name: transformers


LoViM Model Card

Model details

Model type: LoViM is an open-source Vision-Languagde model trained by initializing from InstructBLIP and alignment with Vicuna on multimodal instruction-finetuning data. It composes of an EVA-CLIP vision encoder, a Q-Former, a projection layer and an auto-regressive language model, based on the decoder only transformer architecture.

Model date: LoViM_FlanT5 was trained in July 2023.

Paper or resources for more information: https://gordonhu608.github.io/lovim/

License: BSD 3-Clause License

Where to send questions or comments about the model: https://github.com/mlpc-ucsd/LoViM

Intended use

Primary intended uses: The primary use of LoViM FlanT5 is for commercial use on large multimodal models.

Primary intended users: The primary intended users of this model is for commercial companies in computer vision, natural language processing, machine learning, and artificial intelligence.

Training dataset

Pre-train data: 558K filtered image-text pairs from LAION,CC-3M, and SBU. Selected by LLaVA.

Instruction-finetuning data: COCO-Caption, TextCaps, VQAv2, OKVQA, A-OKVQA, LLaVA-150K, OCR-VQA.

Evaluation dataset

For zero-shot evaluation on general image task, we selected Nocaps, Flickr30K, VizWiz, Visual Spaial Reasoning (VSR), IconQA, Visual Dialog, ScienceQA, MSRVTT QA, TextVQA and Hateful Memes.

For zero-shot evaluation on text-rich image OCR task, we selected ST-VQA, OCR-VQA, Text-VQA, and Doc-VQA.

More detials are in our github, https://github.com/mlpc-ucsd/LoViM