StanfordAIMI / CheXOne

📝 Paper • 🤗 Hugging Face • 🧩 Github • 🪄 Project

✨ Key Features:

• Reasoning Capability : Produces explicit reasoning traces alongside final answers.

• Multi-Task Support : Supports Visual Question Answering (VQA), Report Generation, and Visual Grounding.

• Resident-Level Report Drafting : Matches or outperforms resident-drafted reports in 50% of cases.

• Two Inference Modes
  • Reasoning Mode : Higher performance with explicit reasoning traces.
  • Instruct Mode : Faster inference without reasoning traces.

🎬 Get Started

CheXOne is post-trained on Qwen2.5VL-3B-Instruct model, which 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 accelerate

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(
    "StanfordAIMI/CheXOne", torch_dtype="auto", device_map="auto"
)

# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image scenarios.
# model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
#     "StanfordAIMI/CheXOne",
#     torch_dtype=torch.bfloat16,
#     attn_implementation="flash_attention_2",
#     device_map="auto",
# )

# default processer
processor = AutoProcessor.from_pretrained("StanfordAIMI/CheXOne")

# The default range for the number of visual tokens per image in the model is 4-16384.
# We recommand to set max_pixels=512*512 to align with the training setting.
# min_pixels = 256*28*28
# max_pixels = 512*512
# processor = AutoProcessor.from_pretrained("StanfordAIMI/CheXOne", min_pixels=min_pixels, max_pixels=max_pixels)

# Inference Mode: Reasoning
messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://github.com/YBZh/CheXOne/blob/main/asset/cxr.jpg",
            },
            {"type": "text", "text": "Write an example findings section for the CXR. Please reason step by step, and put your final answer within \\boxed{{}}."},
        ],
    }
]

# Inference Mode: Instruct
# messages = [
#     {
#         "role": "user",
#         "content": [
#             {
#                 "type": "image",
#                 "image": "https://github.com/YBZh/CheXOne/blob/main/asset/cxr.jpg",
#             },
#             {"type": "text", "text": "Write an example findings section for the CXR."},
#         ],
#     }
# ]

# 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=1024)
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)

# Messages containing multiple images and a text query
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "https://github.com/YBZh/CheXOne/blob/main/asset/cxr.jpg"},
            {"type": "image", "image": "https://github.com/YBZh/CheXOne/blob/main/asset/cxr_lateral.jpg"},
            {"type": "text", "text": "Write an example findings section for the CXR. Please reason step by step, and put your final answer within \\boxed{{}}."},
        ],
    }
]

# 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
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

@article{xx,
  title={xx},
  author={Cxxx},
  journal={xx},
  url={xx},
  year={xx}
}