TIGeR: Tool-Integrated Geometric Reasoning in Vision-Language Models for Robotics

Usage

Environment Requirements

pip install -r requirements.txt

Configuration

Before using the model, you need to update the configuration file glm4v_tisr_full_inference.yaml:

1. Update media_dir to your image directory:

   media_dir: /path/to/your/images
   

2. Update the image path in example_usage.py:

   image_paths = ["/path/to/your/image.jpg"]  # Replace with actual image path
   

Basic Usage

import sys
from llamafactory.chat.chat_model import ChatModel

# Load model using LLaMA-Factory ChatModel
config_file = "glm4v_tisr_full_inference.yaml"

# Simulate command line arguments
original_argv = sys.argv.copy()
sys.argv = [sys.argv[0], config_file]

try:
    chat_model = ChatModel()
finally:
    # Restore original command line arguments
    sys.argv = original_argv

# Prepare input
image_paths = ["/path/to/your/image.jpg"]  # Replace with actual image path
question = "Two points are circled on the image, labeled by A and B beside each circle. Which point is closer to the camera? Select from the following choices.\n(A) A is closer\n(B) B is closer"

# Prepare messages
messages = [
    {
        "role": "user",
        "content": question
    }
]

# Get model response
response = chat_model.chat(messages, images=image_paths)
assistant_texts = []

for resp in response:
    try:
        assistant_texts.append(resp.response_text)
    except Exception:
        assistant_texts.append(str(resp))

response_text = "\n".join(assistant_texts)
print(response_text)

Citation

If you use this model, please cite:

@misc{2510.07181,
  Author = {Yi Han and Cheng Chi and Enshen Zhou and Shanyu Rong and Jingkun An and Pengwei Wang and Zhongyuan Wang and Lu Sheng and Shanghang Zhang},
  Title = {TIGeR: Tool-Integrated Geometric Reasoning in Vision-Language Models for Robotics},
  Year = {2025},
  Eprint = {arXiv:2510.07181},
}