SWIM-7B

Paper | GitHub | NL-Refer Dataset

This repository contains the baseline model for See What I Mean: Aligning Vision and Language Representations for Video Fine-grained Object Understanding.

Model Summary

This repository contains the baseline model SWIM-7B. This model is fine-tuned from Qwen2.5-VL model with SIGLIP vision encoder and Qwen2.5-7B-Instruct large language model.

SWIM shares a same architecture with Qwen2.5-VL, You can directly replace "Qwen/Qwen2.5-VL-7B-Instruct" to "BBBBCHAN/SWIM-7B" to get fine-grained object understanding with nature language.

Quick Start

Here we provide a quick run script for SWIM-7B adopted from Qwen2.5-VL.

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

# default: Load the model on the available device(s)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "BBBBCHAN/SWIM-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 = Qwen2_5_VLForConditionalGeneration.from_pretrained(
#     "BBBBCHAN/SWIM-7B",
#     torch_dtype=torch.bfloat16,
#     attn_implementation="flash_attention_2",
#     device_map="auto",
# )

# default processer
processor = AutoProcessor.from_pretrained("BBBBCHAN/SWIM-7B")

# 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 range of 256-1280, to balance performance and cost.
# min_pixels = 256*28*28
# max_pixels = 1280*28*28
# processor = AutoProcessor.from_pretrained("BBBBCHAN/SWIM-7B", min_pixels=min_pixels, max_pixels=max_pixels)


# Messages containing a local video path and a text query
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "video",
                "video": "file:///path/to/video1.mp4",
                "max_pixels": 360 * 420,
                "fps": 1.0,
            },
            {"type": "text", "text": "Describe this video."},
        ],
    }
]

#In Qwen 2.5 VL, frame rate information is also input into the model to align with absolute time.
# Preparation for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs, video_kwargs = process_vision_info(messages, return_video_kwargs=True)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    fps=fps,
    padding=True,
    return_tensors="pt",
    **video_kwargs,
)
inputs = inputs.to("cuda")

# Inference
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)

Citation

If you find our repo useful for your research, please consider citing our paper:

@inproceedings{sun2026swim,
  title     = {See What I Mean: Aligning Vision and Language Representations
               for Video Fine-grained Object Understanding},
  author    = {Sun, Boyuan and Yin, Bowen and Li, Yuanming and Wei, Xihan and Hou, Qibin},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026}
}