Home

Awesome

<!-- Copyright 2023 Cruise LLC -->

šŸŽÆ [CVPR2024] ViP-LLaVA: Making Large Multimodal Models Understand Arbitrary Visual Prompts

Mu Cai, Haotian Liu, Siva Karthik Mustikovela, Gregory P. Meyer, Yuning Chai, Dennis Park, Yong Jae Lee

Approach: [Project Page] [Demo] [Model Zoo] [Paper]

ViP-Bench: [Download Dataset] [LeaderBoard] [Evaluation Server]

<img src="images/vip-llava_arch.png" width="80%">

Release

Code License Data License Usage and License Notices: The data and checkpoint is intended and licensed for research use only. They are also restricted to uses that follow the license agreement of LLaMA, Vicuna and GPT-4. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes.

Contents

Install

If you are not using Linux, do NOT proceed, see instructions for macOS and Windows.

  1. Clone this repository and navigate to ViP-LLaVA folder
cd ViP-LLaVA
  1. Install Package
conda create -n vip-llava python=3.10 -y
conda activate vip-llava
pip install --upgrade pip  # enable PEP 660 support
pip install -e .
  1. Install additional packages for training cases
pip install -e ".[train]"
pip install flash-attn --no-build-isolation

Quick Start With HuggingFace

<details> <summary>Example Code</summary>
from llava.model.builder import load_pretrained_model
from llava.mm_utils import get_model_name_from_path
from llava.eval.run_llava import eval_model

model_path = "mucai/vip-llava-7b"
prompt = "What is shown within the pointed region?"
image_file = "https://pages.cs.wisc.edu/~mucai/man-cross-street.jpg"

args = type('Args', (), {
    "model_path": model_path,
    "model_name": get_model_name_from_path(model_path),
    "query": prompt,
    "image_file": image_file,
    "conv_mode": None, "model_base": None, "temperature": 0.2, "top_p": None, "num_beams": 1, "max_new_tokens": 512, "sep": ",",
})()

eval_model(args)

Check out the details wth the load_pretrained_model function in llava/model/builder.py.

You can also use the eval_model function in llava/eval/run_llava.py to get the output easily. By doing so, you can use this code on Colab directly after downloading this repository.

</details>

ViP-LLaVA Weights

Please check out our Model Zoo for all public ViP-LLaVA checkpoints, and the instructions of how to use the weights.

Demo

To run our demo, you need to prepare LLaVA checkpoints locally. Please follow the instructions here to download the checkpoints.

Gradio Web UI

To launch a Gradio demo locally, please run the following commands one by one. If you plan to launch multiple model workers to compare between different checkpoints, you only need to launch the controller and the web server ONCE.

flowchart BT
    %% Declare Nodes
    gws("Gradio (UI Server)")
    c("Controller (API Server):<br/>PORT: 10000")
    mw7b("Model Worker:<br/>vip-llava-7b<br/>PORT: 40000")
    mw13b("Model Worker:<br/>vip-llava-13b<br/>PORT: 40001")

    %% Declare Styles
    classDef data fill:#3af,stroke:#48a,stroke-width:2px,color:#444
    classDef success fill:#8f8,stroke:#0a0,stroke-width:2px,color:#444
    classDef failure fill:#f88,stroke:#f00,stroke-width:2px,color:#444

    %% Assign Styles
    class id,od data;
    class cimg,cs_s,scsim_s success;
    class ncimg,cs_f,scsim_f failure;

    subgraph Demo Connections
        direction BT
        c<-->gws
        
        mw7b<-->c
        mw13b<-->c
    end

Launch a controller

python -m llava.serve.controller --host 0.0.0.0 --port 10000

Launch a gradio web server.

python -m llava.serve.gradio_web_server --controller http://localhost:10000 --model-list-mode reload

You just launched the Gradio web interface. Now, you can open the web interface with the URL printed on the screen. You may notice that there is no model in the model list. Do not worry, as we have not launched any model worker yet. It will be automatically updated when you launch a model worker.

Launch a model worker

This is the actual worker that performs the inference on the GPU. Each worker is responsible for a single model specified in --model-path.

python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:10000 --port 40000 --worker http://localhost:40000 --model-path mucai/vip-llava-13b

Wait until the process finishes loading the model and you see "Uvicorn running on ...". Now, refresh your Gradio web UI, and you will see the model you just launched in the model list.

You can launch as many workers as you want, and compare between different model checkpoints in the same Gradio interface. Please keep the --controller the same, and modify the --port and --worker to a different port number for each worker.

python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:10000 --port <different from 40000, say 40001> --worker http://localhost:<change accordingly, i.e. 40001> --model-path <ckpt2>

If you are using an Apple device with an M1 or M2 chip, you can specify the mps device by using the --device flag: --device mps.

Launch a model worker (Multiple GPUs, when GPU VRAM <= 24GB)

If the VRAM of your GPU is less than 24GB (e.g., RTX 3090, RTX 4090, etc.), you may try running it with multiple GPUs. Our latest code base will automatically try to use multiple GPUs if you have more than one GPU. You can specify which GPUs to use with CUDA_VISIBLE_DEVICES. Below is an example of running with the first two GPUs.

CUDA_VISIBLE_DEVICES=0,1 python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:10000 --port 40000 --worker http://localhost:40000 --model-path mucai/vip-llava-13b

Launch a model worker (4-bit, 8-bit inference, quantized)

You can launch the model worker with quantized bits (4-bit, 8-bit), which allows you to run the inference with reduced GPU memory footprint, potentially allowing you to run on a GPU with as few as 12GB VRAM. Note that inference with quantized bits may not be as accurate as the full-precision model. Simply append --load-4bit or --load-8bit to the model worker command that you are executing. Below is an example of running with 4-bit quantization.

python -m llava.serve.model_worker --host 0.0.0.0 --controller http://localhost:10000 --port 40000 --worker http://localhost:40000 --model-path mucai/vip-llava-13b --load-4bit

CLI Inference

Chat about images using ViP-LLaVA without the need of Gradio interface. It also supports multiple GPUs, 4-bit and 8-bit quantized inference. With 4-bit quantization, for our ViP-LLaVA-7B, it uses less than 8GB VRAM on a single GPU.

python -m llava.serve.cli \
    --model-path mucai/vip-llava-7b \
    --image-file "https://pages.cs.wisc.edu/~mucai/man-cross-street.jpg" \
    --load-4bit

Or use the bounding box

python -m llava.serve.cli_vip --model-path ./checkpoints/vip-llava-7b --image-file "https://pages.cs.wisc.edu/~mucai/example_styletransfer.png"  --bbox=100,200,200,300
<img src="images/demo_cli.gif" width="70%">

Train

ViP-LLaVA training consists of three stages: (1) feature alignment stage: use our 558K subset of the LAION-CC-SBU dataset to connect a frozen pretrained vision encoder to a frozen LLM; (2) visual instruction tuning stage: 665K image-level instruction data from LLaVA-1.5 and 520K region-level instruction data using visual prompts. (3) finetuning on GPT-4V data.

LLaVA is trained on 8 A100 GPUs with 80GB memory. To train on fewer GPUs, you can reduce the per_device_train_batch_size and increase the gradient_accumulation_steps accordingly. Always keep the global batch size the same: per_device_train_batch_size x gradient_accumulation_steps x num_gpus.

Hyperparameters

We use a similar set of hyperparameters as Vicuna in finetuning. Both hyperparameters used in pretraining and finetuning are provided below.

  1. Pretraining
HyperparameterGlobal Batch SizeLearning rateEpochsMax lengthWeight decay
ViP-LLaVA-13B2561e-3120480
  1. Finetuning
HyperparameterGlobal Batch SizeLearning rateEpochsMax lengthWeight decay
ViP-LLaVA-13B1282e-5120480

Download Vicuna checkpoints (automatically)

Our base model Vicuna v1.5, which is an instruction-tuned chatbot, will be downloaded automatically when you run our provided training scripts. No action is needed.

Pretrain (feature alignment)

Please download the 558K subset of the LAION-CC-SBU dataset with BLIP captions we use in the paper here.

Pretrain takes around 5.5 hours for ViP-LLaVA-13B on 8x A100 (80G). It takes around 3.5 hours for ViP-LLaVA-7B.

Training script with DeepSpeed ZeRO-2: pretrain.sh.

<details> <summary>Pretraining takes around 20 hours for LLaVA-7B on 8x V100 (32G)</summary>

We provide training script with DeepSpeed here. Tips:

</details>

Visual Instruction Tuning

  1. Prepare data

Please download the annotations of our instruction tuning data in stage 2 and stage 3. and download the images from constituting datasets:

After downloading all of them, organize the data as follows in ./playground/data,

ā”œā”€ā”€ flickr30k-images
ā”œā”€ā”€ v7w
ā”œā”€ā”€ vcr1images
ā”œā”€ā”€ coco
ā”‚   ā””ā”€ā”€ train2017
ā”œā”€ā”€ gqa
ā”‚   ā””ā”€ā”€ images
ā”œā”€ā”€ ocr_vqa
ā”‚   ā””ā”€ā”€ images
ā”œā”€ā”€ textvqa
ā”‚   ā””ā”€ā”€ train_images
ā””ā”€ā”€ vg
    ā”œā”€ā”€ VG_100K
    ā””ā”€ā”€ VG_100K_2
  1. Start training!

You may download our pretrained projectors in Model Zoo. It is not recommended to use legacy projectors, as they may be trained with a different version of the codebase, and if any option is off, the model will not function/train as we expected.

Visual instruction tuning takes around 40 hours for LLaVA-v1.5-13B on 8x A100 (80G). It takes around 20 hours for LLaVA-v1.5-7B on 8x A100 (40G).

Training script with DeepSpeed ZeRO-2: finetune_stage2.sh. If you further want to use GPT-4V data to enhance the chatting capability, see the training script for stage 3 with DeepSpeed ZeRO-2: finetune_stage3.sh

If you do not have enough GPU memory:

If you are interested in finetuning LLaVA model to your own task/data, please check out Finetune_Custom_Data.md怂

Evaluation

ViP-LLaVA is both evaluate on the 4 academic region-level benchmark and the newly proposed ViP-Bench.

See Evaluation.md.

Citation

If you find ViP-LLaVA useful for your research and applications, please cite using this BibTeX:


@inproceedings{cai2024vipllava,
  author      = {Cai, Mu and Liu, Haotian and Mustikovela,  Siva Karthik and Meyer, Gregory P. and Chai, Yuning and Park, Dennis and Lee, Yong Jae},
  title       = {Making Large Multimodal Models Understand Arbitrary Visual Prompts},
  booktitle   = {IEEE Conference on Computer Vision and Pattern Recognition},
  year        = {2024}
}

Acknowledgement