LAVIS - A Library for Language-Vision Intelligence

What's New: 🎉

• [Model Release] Jan 2023, released implementation of BLIP-2
  Paper, Project Page, , > A generic and efficient pre-training strategy that easily harvests development of pretrained vision models and large language models (LLMs) for vision-language pretraining. BLIP-2 beats Flamingo on zero-shot VQAv2 (65.0 vs 56.3), establishing new state-of-the-art on zero-shot captioning (on NoCaps 121.6 CIDEr score vs previous best 113.2). In addition, equipped with powerful LLMs (e.g. OPT, FlanT5), BLIP-2 also unlocks the new zero-shot instructed vision-to-language generation capabilities for various interesting applications!
• Jan 2023, LAVIS is now available on PyPI for installation!
• [Model Release] Dec 2022, released implementation of Img2prompt-VQA
  Paper, Project Page, > A plug-and-play module that enables off-the-shelf use of Large Language Models (LLMs) for visual question answering (VQA). Img2Prompt-VQA surpasses Flamingo on zero-shot VQA on VQAv2 (61.9 vs 56.3), while in contrast requiring no end-to-end training!
• [Model Release] Oct 2022, released implementation of PNP-VQA (EMNLP Findings 2022, "Plug-and-Play VQA: Zero-shot VQA by Conjoining Large Pretrained Models with Zero Training", by Anthony T.M.H. et al),
  Paper, Project Page, ) > A modular zero-shot VQA framework that requires no PLMs training, achieving SoTA zero-shot VQA performance.

Table of Contents

• Introduction
• Installation
• Getting Started
  • Model Zoo
  • Image Captioning
  • Visual question answering (VQA)
  • Unified Feature Extraction Interface
  • Load Datasets
• Jupyter Notebook Examples
• Resources and Tools
• Documentations
• Ethical and Responsible Use
• Technical Report and Citing LAVIS
• License

Introduction

LAVIS is a Python deep learning library for LAnguage-and-VISion intelligence research and applications. This library aims to provide engineers and researchers with a one-stop solution to rapidly develop models for their specific multimodal scenarios, and benchmark them across standard and customized datasets. It features a unified interface design to access

• 10+ tasks (retrieval, captioning, visual question answering, multimodal classification etc.);
• 20+ datasets (COCO, Flickr, Nocaps, Conceptual Commons, SBU, etc.);
• 30+ pretrained weights of state-of-the-art foundation language-vision models and their task-specific adaptations, including ALBEF, BLIP, ALPRO, CLIP.

Key features of LAVIS include:

• Unified and Modular Interface: facilitating to easily leverage and repurpose existing modules (datasets, models, preprocessors), also to add new modules.

• Easy Off-the-shelf Inference and Feature Extraction: readily available pre-trained models let you take advantage of state-of-the-art multimodal understanding and generation capabilities on your own data.

• Reproducible Model Zoo and Training Recipes: easily replicate and extend state-of-the-art models on existing and new tasks.

• Dataset Zoo and Automatic Downloading Tools: it can be a hassle to prepare the many language-vision datasets. LAVIS provides automatic downloading scripts to help prepare a large variety of datasets and their annotations.

The following table shows the supported tasks, datasets and models in our library. This is a continuing effort and we are working on further growing the list.

Installation

1. (Optional) Creating conda environment

conda create -n lavis python=3.8
conda activate lavis

2. install from PyPI

pip install salesforce-lavis

3. Or, for development, you may build from source

git clone https://github.com/salesforce/LAVIS.git
cd LAVIS
pip install -e .

Getting Started

Model Zoo

Model zoo summarizes supported models in LAVIS, to view:

from lavis.models import model_zoo
print(model_zoo)
# ==================================================
# Architectures                  Types
# ==================================================
# albef_classification           ve
# albef_feature_extractor        base
# albef_nlvr                     nlvr
# albef_pretrain                 base
# albef_retrieval                coco, flickr
# albef_vqa                      vqav2
# alpro_qa                       msrvtt, msvd
# alpro_retrieval                msrvtt, didemo
# blip_caption                   base_coco, large_coco
# blip_classification            base
# blip_feature_extractor         base
# blip_nlvr                      nlvr
# blip_pretrain                  base
# blip_retrieval                 coco, flickr
# blip_vqa                       vqav2, okvqa, aokvqa
# clip_feature_extractor         ViT-B-32, ViT-B-16, ViT-L-14, ViT-L-14-336, RN50
# clip                           ViT-B-32, ViT-B-16, ViT-L-14, ViT-L-14-336, RN50
# gpt_dialogue                   base

Let’s see how to use models in LAVIS to perform inference on example data. We first load a sample image from local.

import torch
from PIL import Image
# setup device to use
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# load sample image
raw_image = Image.open("docs/_static/merlion.png").convert("RGB")

This example image shows Merlion park (source), a landmark in Singapore.

Image Captioning

In this example, we use the BLIP model to generate a caption for the image. To make inference even easier, we also associate each pre-trained model with its preprocessors (transforms), accessed via load_model_and_preprocess().

import torch
from lavis.models import load_model_and_preprocess
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# loads BLIP caption base model, with finetuned checkpoints on MSCOCO captioning dataset.
# this also loads the associated image processors
model, vis_processors, _ = load_model_and_preprocess(name="blip_caption", model_type="base_coco", is_eval=True, device=device)
# preprocess the image
# vis_processors stores image transforms for "train" and "eval" (validation / testing / inference)
image = vis_processors["eval"](raw_image).unsqueeze(0).to(device)
# generate caption
model.generate({"image": image})
# ['a large fountain spewing water into the air']

Visual question answering (VQA)

BLIP model is able to answer free-form questions about images in natural language. To access the VQA model, simply replace the name and model_type arguments passed to load_model_and_preprocess().

from lavis.models import load_model_and_preprocess
model, vis_processors, txt_processors = load_model_and_preprocess(name="blip_vqa", model_type="vqav2", is_eval=True, device=device)
# ask a random question.
question = "Which city is this photo taken?"
image = vis_processors["eval"](raw_image).unsqueeze(0).to(device)
question = txt_processors["eval"](question)
model.predict_answers(samples={"image": image, "text_input": question}, inference_method="generate")
# ['singapore']

Unified Feature Extraction Interface

LAVIS provides a unified interface to extract features from each architecture. To extract features, we load the feature extractor variants of each model. The multimodal feature can be used for multimodal classification. The low-dimensional unimodal features can be used to compute cross-modal similarity.

from lavis.models import load_model_and_preprocess
model, vis_processors, txt_processors = load_model_and_preprocess(name="blip_feature_extractor", model_type="base", is_eval=True, device=device)
caption = "a large fountain spewing water into the air"
image = vis_processors["eval"](raw_image).unsqueeze(0).to(device)
text_input = txt_processors["eval"](caption)
sample = {"image": image, "text_input": [text_input]}

features_multimodal = model.extract_features(sample)
print(features_multimodal.multimodal_embeds.shape)
# torch.Size([1, 12, 768]), use features_multimodal[:,0,:] for multimodal classification tasks

features_image = model.extract_features(sample, mode="image")
features_text = model.extract_features(sample, mode="text")
print(features_image.image_embeds.shape)
# torch.Size([1, 197, 768])
print(features_text.text_embeds.shape)
# torch.Size([1, 12, 768])

# low-dimensional projected features
print(features_image.image_embeds_proj.shape)
# torch.Size([1, 197, 256])
print(features_text.text_embeds_proj.shape)
# torch.Size([1, 12, 256])
similarity = features_image.image_embeds_proj[:,0,:] @ features_text.text_embeds_proj[:,0,:].t()
print(similarity)
# tensor([[0.2622]])

Load Datasets

LAVIS inherently supports a wide variety of common language-vision datasets by providing automatic download tools to help download and organize these datasets. After downloading, to load the datasets, use the following code:

from lavis.datasets.builders import dataset_zoo
dataset_names = dataset_zoo.get_names()
print(dataset_names)
# ['aok_vqa', 'coco_caption', 'coco_retrieval', 'coco_vqa', 'conceptual_caption_12m',
#  'conceptual_caption_3m', 'didemo_retrieval', 'flickr30k', 'imagenet', 'laion2B_multi',
#  'msrvtt_caption', 'msrvtt_qa', 'msrvtt_retrieval', 'msvd_caption', 'msvd_qa', 'nlvr',
#  'nocaps', 'ok_vqa', 'sbu_caption', 'snli_ve', 'vatex_caption', 'vg_caption', 'vg_vqa']

After downloading the images, we can use load_dataset() to obtain the dataset.

from lavis.datasets.builders import load_dataset
coco_dataset = load_dataset("coco_caption")
print(coco_dataset.keys())
# dict_keys(['train', 'val', 'test'])
print(len(coco_dataset["train"]))
# 566747
print(coco_dataset["train"][0])
# {'image': ,
#  'text_input': 'A woman wearing a net on her head cutting a cake. ',
#  'image_id': 0}

If you already host a local copy of the dataset, you can pass in the vis_path argument to change the default location to load images.

coco_dataset = load_dataset("coco_caption", vis_path=YOUR_LOCAL_PATH)

Jupyter Notebook Examples

See examples for more inference examples, e.g. captioning, feature extraction, VQA, GradCam, zeros-shot classification.

Resources and Tools

• Benchmarks: see Benchmark for instructions to evaluate and train supported models.
• Dataset Download and Browsing: see Dataset Download for instructions and automatic tools on download common language-vision datasets.
• GUI Demo: to run the demo locally, run bash run_scripts/run_demo.sh and then follow the instruction on the prompts to view in browser. A web demo is coming soon.

Documentations

For more details and advanced usages, please refer to documentation.

Ethical and Responsible Use

We note that models in LAVIS provide no guarantees on their multimodal abilities; incorrect or biased predictions may be observed. In particular, the datasets and pretrained models utilized in LAVIS may contain socioeconomic biases which could result in misclassification and other unwanted behaviors such as offensive or inappropriate speech. We strongly recommend that users review the pre-trained models and overall system in LAVIS before practical adoption. We plan to improve the library by investigating and mitigating these potential biases and inappropriate behaviors in the future.

Technical Report and Citing LAVIS

You can find more details in our technical report.

If you're using LAVIS in your research or applications, please cite using this BibTeX:

@misc{li2022lavis,
      title={LAVIS: A Library for Language-Vision Intelligence}, 
      author={Dongxu Li and Junnan Li and Hung Le and Guangsen Wang and Silvio Savarese and Steven C. H. Hoi},
      year={2022},
      eprint={2209.09019},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Contact us

If you have any questions, comments or suggestions, please do not hesitate to contact us at lavis@salesforce.com.

License

BSD 3-Clause License