BGE-VL-MLLM-S2

News

2024-3-4 🚀🚀 We have released the BGE-VL-MLLM models on Huggingface: BGE-VL-MLLM-S1 and BGE-VL-MLLM-S2. BGE-VL-MLLM-S1 is trained exclusively on our MegaPairs dataset, achieving outstanding performance in composed image retrieval, with an 8.1% improvement on the CIRCO benchmark (mAP@5) over the previous state-of-the-art. BGE-VL-MLLM-S2 builds on BGE-VL-MLLM-S1 with an additional epoch of fine-tuning on the MMEB benchmark training set, delivering enhanced performance across a broader range of multimodal embedding tasks.

2024-12-27 🚀🚀 BGE-VL-CLIP models are released on Huggingface: BGE-VL-base and BGE-VL-large.

2024-12-19 🎉🎉 Release our paper: MegaPairs: Massive Data Synthesis For Universal Multimodal Retrieval.

Release Plan

• Paper
• BGE-VL-base and BGE-VL-large models
• BGE-VL-MLLM model
• MegaPairs Dataset
• Evaluation code
• Fine-tuning code

Introduction

In this work, we introduce MegaPairs, a novel data synthesis method that leverages open-domain images to create heterogeneous KNN triplets for universal multimodal retrieval. Our MegaPairs dataset contains over 26 million triplets, and we have trained a series of multimodal retrieval models, BGE-VL, including BGE-VL-CLIP (base and large) and BGE-VL-MLLM.

BGE-VL achieve state-of-the-art performance on four popular zero-shot composed image retrieval benchmarks and the massive multimodal embedding benchmark (MMEB). Extensive experiments demonstrate the efficiency, scalability, and generalization features of MegaPairs. Please refer to our paper for more details.

Model Usage

Using Sentence Transformers

Install Sentence Transformers:

pip install sentence_transformers[image]

This model supports text, image, and combined text+image inputs. Images must be resized to 512x512, as the model was trained with that resolution. Pass a prompt with a task instruction to format inputs as queries, or omit it to format as candidates.

import torch
from sentence_transformers import SentenceTransformer
from PIL import Image

model = SentenceTransformer("BAAI/BGE-VL-MLLM-S2", model_kwargs={"torch_dtype": torch.float16})

# Composed image retrieval: text + image query
query_img = Image.open("./assets/cir_query.png").resize((512, 512))
query_emb = model.encode(
    {"text": "Make the background dark, as if the camera has taken the photo at night", "image": query_img},
    prompt="Retrieve the target image that best meets the combined criteria by using both the provided image and the image retrieval instructions: ",
)

# Image-only candidates (no prompt = candidate format)
candidate_imgs = [
    Image.open("./assets/cir_candi_1.png").resize((512, 512)),
    Image.open("./assets/cir_candi_2.png").resize((512, 512)),
]
candidate_embs = model.encode(candidate_imgs, batch_size=1)
print(candidate_embs.shape)
# (2, 4096)

similarities = model.similarity(query_emb, candidate_embs)
print(similarities)
# tensor([[0.4158, 0.2188]], dtype=torch.float16)

1. BGE-VL-CLIP Models

You can easily use BGE-VL-CLIP models based on transformers

import torch
from transformers import AutoModel

MODEL_NAME = "BAAI/BGE-VL-base" # or "BAAI/BGE-VL-large"

model = AutoModel.from_pretrained(MODEL_NAME, trust_remote_code=True) # You must set trust_remote_code=True
model.set_processor(MODEL_NAME)
model.eval()

with torch.no_grad():
    query = model.encode(
        images = "./assets/cir_query.png", 
        text = "Make the background dark, as if the camera has taken the photo at night"
    )

    candidates = model.encode(
        images = ["./assets/cir_candi_1.png", "./assets/cir_candi_2.png"]
    )
    
    scores = query @ candidates.T
print(scores)

See the demo for a complete example of using BGE-VL for multimodel retrieval.

2. BGE-VL-MLLM Models

import torch
from transformers import AutoModel
from PIL import Image

MODEL_NAME= "BAAI/BGE-VL-MLLM-S1"

model = AutoModel.from_pretrained(MODEL_NAME, trust_remote_code=True)
model.eval()
model.cuda()

with torch.no_grad():
    model.set_processor(MODEL_NAME)

    query_inputs = model.data_process(
        text="Make the background dark, as if the camera has taken the photo at night", 
        images="./assets/cir_query.png",
        q_or_c="q",
        task_instruction="Retrieve the target image that best meets the combined criteria by using both the provided image and the image retrieval instructions: "
    )

    candidate_inputs = model.data_process(
        images=["./assets/cir_candi_1.png", "./assets/cir_candi_2.png"],
        q_or_c="c",
    )

    query_embs = model(**query_inputs, output_hidden_states=True)[:, -1, :]
    candi_embs = model(**candidate_inputs, output_hidden_states=True)[:, -1, :]
    
    query_embs = torch.nn.functional.normalize(query_embs, dim=-1)
    candi_embs = torch.nn.functional.normalize(candi_embs, dim=-1)

    scores = torch.matmul(query_embs, candi_embs.T)
print(scores)

Model Performance

Zero-Shot Composed Image Retrieval

BGE-VL sets a new performance benchmark in zero-shot composed image retrieval tasks. On the CIRCO benchmark, our BGE-VL-base model, with only 149 million parameters, surpasses all previous models, including those with 50 times more parameters. Additionally, BGE-VL-MLLM achieves an 8.1% improvement over the previous state-of-the-art model.

Zero-Shot Performance on MMEB

BGE-VL-MLLM achieves state-of-the-art zero-shot performance on the Massive Multimodal Embedding Benchmark (MMEB), despite being trained only on the ImageText-to-Image paradigm. This demonstrates the excellent generalization capability of MegaPairs for multimodal embedding.

Fine-Tuning Performance on MMEB

After fine-tuning on downstream tasks, BGE-VL-MLLM maintains its leading performance. Notably, it surpasses the previous state-of-the-art by 7.1% on the MMEB out-of-distribution (OOD) set. These results demonstrate the robust generalization capability of BGE-VL-MLLM and highlight the potential of MegaPairs as foundational training data for universal multimodal embedding.

Performance Scaling

MegaPairs showcases scalability: BGE-VL-base improves as training data increases. It also demonstrates efficiency: with just 0.5M training samples, BGE-VL-base significantly outperforms MagicLens, which uses the same CLIP-base backbone and was trained on 36.7M samples.

License

The annotations for MegaPairs and the BGE-VL models are released under the MIT License. The images in MegaPairs originate from the Recap-Datacomp, which is released under the CC BY 4.0 license.

Citation

If you find this repository useful, please consider giving a star ⭐ and citation

@article{zhou2024megapairs,
  title={MegaPairs: Massive Data Synthesis For Universal Multimodal Retrieval},
  author={Zhou, Junjie and Liu, Zheng and Liu, Ze and Xiao, Shitao and Wang, Yueze and Zhao, Bo and Zhang, Chen Jason and Lian, Defu and Xiong, Yongping},
  journal={arXiv preprint arXiv:2412.14475},
  year={2024}
}