Discrete token-based multimodal large language models (MLLMs), such as AnyGPT and MIO, integrate diverse modalities into an autoregressive framework by discretizing modality inputs into tokens compatible with language models. Unlike encoder-based approaches, such as LLaVA and Flamingo, which utilize pretrained modality-specific encoders, discrete token-based MLLMs simultaneously learn modality token representations and their alignment with the language, yet are exclusively trained on modality-text paired datasets without additional unimodal training. We identify a structural limitation inherent in this training paradigm, termed text-centric bias, defined as an over-reliance on the textual context that restricts intrinsic modality understanding. To systematically analyze the existence of this bias, we propose an analytical framework involving external perplexity-based and internal neuron-level analyses. Furthermore, to verify whether the bias originates from the paired-only training paradigm, we introduce an analytical methodology named Monotune, which is a simple unimodal training stage. Our analyses demonstrate that minimal exposure to unimodal data effectively mitigates text-centric bias, providing empirical evidence that the bias is fundamentally induced by the paired-only training strategy. Through comprehensive downstream task evaluations, we further reveal that this structural bias meaningfully affects real-world multimodal task performance, particularly under limited textual contexts. Our findings highlight a fundamental limitation in current discrete token-based MLLM training paradigms and suggest directions for future multimodal training strategies. Our code and experiments are available at https://github.com/41312432/Monotune

Omnidirectional vision systems provide a 360-degree panoramic view, enabling full environmental awareness in various fields, such as Advanced Driver Assistance Systems (ADAS) and Virtual Reality (VR). Existing omnidirectional stitching methods rely on a single specialized 360-degree camera. However, due to hardware limitations such as high mounting heights and blind spots, adapting these methods to vehicles of varying sizes and geometries is challenging. These challenges include limited generalizability due to the reliance on predefined stitching regions for fixed camera arrays, performance degradation from distance parallax leading to large depth differences, and the absence of suitable datasets with ground truth for multi-camera omnidirectional systems. To overcome these challenges, we propose a novel omnidirectional stitching framework and a publicly available dataset tailored for varying distance scenarios with multiple cameras. The framework, referred to as OmniStitch, consists of a Stitching Region Maximization (SRM) module for automatic adaptation to different vehicles with multiple cameras and a Depth-Aware Stitching (DAS) module to handle depth differences caused by distance parallax between cameras. In addition, we create and release an omnidirectional stitching dataset, called GV360, which provides ground truth images that maintain the perspective of the 360-degree FOV, designed explicitly for vehicle-agnostic systems. Extensive evaluations of this dataset demonstrate that our framework outperforms state-of-the-art stitching models, especially in handling varying distance parallax. The proposed dataset and code are publicly available in https://github.com/tngh5004/Omnistitch.