Enhancing Large Vision Language Models with Self-Training on Image Comprehension
We introduce STIC (Self-Training on Image Comprehension) that enhances the understanding and reasoning capabilities of LVLMs through self-generated data. Our experiments across seven benchmarks, including ScienceQA, TextVQA, ChartQA, LLaVA-Bench, MMBench, MM-Vet, and MathVista, demonstrate a notable average accuracy gain of 4.0% by self-training.
May 30, 2024
