RC-Qwen2VL-7B

Arxiv: 📑Region-Level Context-Aware Multimodal Understanding
Models：🤗RC-Qwen2VL-2b | 🤗RC-Qwen2VL-7b
Demos: 🚀Personalized Conversations About Images | 🚀Celebrity Recognition and VQA
Github: https://github.com/hongliang-wei/RC-MLLM

Introduction

RC-Qwen2-VL is a model focused on Region-level Context-aware Multimodal Understanding (RCMU).

• Objective: To solve the key challenge of integrating the visual content of specific objects in an image with their associated text for comprehensive understanding.
• Method: Trained with the RCVIT(Region-level Context-aware Visual Instruction Tuning) method and RCMU dataset, it uses bounding boxes to precisely link visual content with text.
• Performance & Applications: It achieves outstanding performance on RCMU tasks and is successfully applied in advanced scenarios like multimodal RAG and personalized conversation.

Refer to Qwen2-VL for the requirements:

Requirements:

The code of Qwen2-VL has been in the latest Hugging face transformers and we advise you to build from source with command pip install git+https://github.com/huggingface/transformers, or you might encounter the following error:

KeyError: 'qwen2_vl'

Quickstart:

We offer a toolkit to help you handle various types of visual input more conveniently. This includes base64, URLs, and interleaved images and videos. You can install it using the following command:

pip install qwen-vl-utils

Prompt format:

 messages = [
     {
         "role": "user",
         "content": [
             {
                 "type": "image",
                 "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
             },
             {
                 "type": "text",
                 "text": "[1]: The information of the {obj_1} located at <|box_start|>(x1,y1),(x2,y2)<|box_end|> in the image: {info_1}.\n[2]: The information of the {obj_2} located at <|box_start|>(x1,y1),(x2,y2)<|box_end|> in the image: {info_2}......"
             }
     }
 ]

Where the bbox coordinates range from 0 to 1000.

Here we show a code snippet:

from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
# default: Load the model on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(
    "weihongliang/RC-Qwen2VL-7b", torch_dtype="auto", device_map="auto"
)
# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
# model = Qwen2VLForConditionalGeneration.from_pretrained(
#     "weihongliang/RC-Qwen2VL-7b",
#     torch_dtype=torch.bfloat16,
#     attn_implementation="flash_attention_2",
#     device_map="auto",
# )
# default processer
processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
# min_pixels = 256*28*28
# max_pixels = 1280*28*28
# processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels)
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {
                "type": "text",
                "text": "[1]: The information of the man located at <|box_start|>(755,417),(991,885)<|box_end|> in the image: Lily, a 34-year-old marketing executive, resides in downtown Chicago. She completed her MBA at the University of Chicago in 2012 and has a passion for technology and innovation. In her spare time, she enjoys hiking and photography, often capturing stunning landscapes during her outdoor adventures. Currently, Lily is working on a project that focuses on digital marketing strategies, with a vision to integrate artificial intelligence. She also volunteers at a local animal shelter on weekends, promoting animal welfare and seeking to help abandoned pets find their forever homes.\n[2]: The information of the dog located at <|box_start|>(214,424),(583 884)<|box_end|> in the image: Bella is a 2-year-old dog owned by Lily, who emphasizes the joy pets bring to daily life. Bella was brought into the family on her adoption day in April 2022 from a nearby rescue organization. Lily enjoys spending quality time with her, and together they play a variety of games like hide and seek, where Bella loves to dart around furniture and pounce unexpectedly. Bella's preferred activities include watching birds from the window and exploring new boxes or bags that find their way into the house, always keeping Lily entertained with her curious antics.\nAnswer the following question based on the information above and the given image, and provide citations for your response.\nDescribe the image."
            }
    }
]
# Preparation for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)