World to Code: Multi-modal Data Generation via Self-Instructed Compositional Captioning and Filtering
This work addresses the problem of data scarcity for multi-modal AI researchers by providing a novel data generation method, though it is incremental as it builds on existing VLM capabilities.
The paper tackles the scarcity of high-quality multi-modal alignment data by introducing World to Code (W2C), a pipeline that generates synthetic data in Python code format using Vision-Language Models (VLMs) with self-instructed compositional captioning and filtering, resulting in improved performance on visual question answering and visual grounding benchmarks across different VLMs.
Recent advances in Vision-Language Models (VLMs) and the scarcity of high-quality multi-modal alignment data have inspired numerous researches on synthetic VLM data generation. The conventional norm in VLM data construction uses a mixture of specialists in caption and OCR, or stronger VLM APIs and expensive human annotation. In this paper, we present World to Code (W2C), a meticulously curated multi-modal data construction pipeline that organizes the final generation output into a Python code format. The pipeline leverages the VLM itself to extract cross-modal information via different prompts and filter the generated outputs again via a consistency filtering strategy. Experiments have demonstrated the high quality of W2C by improving various existing visual question answering and visual grounding benchmarks across different VLMs. Further analysis also demonstrates that the new code parsing ability of VLMs presents better cross-modal equivalence than the commonly used detail caption ability. Our code is available at https://github.com/foundation-multimodal-models/World2Code.