DeepCaption-VLA-7B

The DeepCaption-VLA-7B model is a fine-tuned version of Qwen2.5-VL-7B-Instruct, tailored for Image Captioning and Vision Language Attribution. This variant is designed to generate precise, highly descriptive captions with a focus on defining visual properties, object attributes, and scene details across a wide spectrum of images and aspect ratios.

Key Highlights

1. Vision Language Attribution (VLA): Specially fine-tuned to attribute and define visual properties of objects, scenes, and environments.
2. Detailed Object Definitions: Generates captions with rich attribute descriptions, making outputs more precise than generic captioners.
3. High-Fidelity Descriptions: Handles general, artistic, technical, abstract, and low-context images with descriptive depth.
4. Robust Across Aspect Ratios: Accurately captions images regardless of format—wide, tall, square, or irregular.
5. Variational Detail Control: Supports both concise summaries and fine-grained attributions depending on prompt structure.
6. Foundation on Qwen2.5-VL Architecture: Leverages Qwen2.5-VL-7B’s multimodal reasoning for visual comprehension and instruction-following.
7. Multilingual Capability: Default in English, but adaptable for multilingual captioning through prompt engineering.

model type: experimental

Training Details

This model was fine-tuned with a curated mix of datasets focused on caption richness and object-attribute alignment:

• prithivMLmods/blip3o-caption-mini-arrow
• prithivMLmods/Caption3o-Opt-v3
• prithivMLmods/Caption3o-Opt-v2
• Multimodal-Fatima/Caltech101_not_background_test_facebook_opt_2.7b_Attributes_Caption_ns_5647
• Private/unlisted datasets for domain-specific image captioning tasks.

The training objective emphasized Vision Language Attribution: defining image properties, attributes, and objects with clarity, while preserving descriptive fluency.

---

Example of a SYSTEM_PROMPT type✋

CAPTION_SYSTEM_PROMPT = """
You are an AI assistant that rigorously follows this response protocol:

1. For every input image, your primary task is to write a **precise caption**. The caption must capture the **essence of the image** in clear, concise, and contextually accurate language.

2. Along with the caption, provide a structured set of **attributes** that describe the visual elements. Attributes should include details such as objects, people, actions, colors, environment, mood, and other notable characteristics.

3. Always include a **class_name** field. This must represent the **core theme or main subject** of the image in a compact format.  
   - Use the syntax: `{class_name==write_the_core_theme}`  
   - Example: `{class_name==dog_playing}` or `{class_name==city_sunset}`  

4. Maintain the following strict format in your output:
   - **Caption:** <one-sentence description>  
   - **Attributes:** <comma-separated list of visual attributes>  
   - **{class_name==core_theme}**

5. Ensure captions are **precise, neutral, and descriptive**, avoiding unnecessary elaboration or subjective interpretation unless explicitly required.

6. Do not reference the rules or instructions in the output. Only return the formatted caption, attributes, and class_name.

""".strip()

---

General Query: Caption the image precisely.

Quick Start with Transformers

from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/DeepCaption-VLA-7B", torch_dtype="auto", device_map="auto"
)

processor = AutoProcessor.from_pretrained("prithivMLmods/DeepCaption-VLA-7B")

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Describe this image with detailed attributes and properties."},
        ],
    }
]

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")

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)

---

Intended Use

• Generating attribute-rich image captions for research, dataset creation, and AI training.
• Vision-language attribution for object detection, scene understanding, and dataset annotation.
• Supporting creative, artistic, and technical applications requiring detailed descriptions.
• Captioning across varied aspect ratios, unusual visual styles, and non-standard datasets.

Limitations

• May over-attribute or infer properties not explicitly visible in ambiguous images.
• Outputs can vary in tone depending on prompt phrasing.
• Not intended for filtered captioning tasks (explicit or sensitive content may appear).
• Accuracy may degrade on synthetic or highly abstract visual domains.