IJCV 2027特刊开启征稿：多模态理解与生成走向统一

TL;DR - The International Journal of Computer Vision (IJCV) has opened a call for papers for a special issue on Multimodal Unified Comprehension and Generation (MUCG), targeting models that jointly handle perception, understanding, reasoning, and generation in one system. It signals that the field's center of gravity is shifting from task-specific multimodal models to unified architectures with closed-loop understanding–generation evaluation.

• Framing: three shifts sought — task-specific → unified models, static perception → interactive intelligence (understanding guides generation and vice versa), isolated metrics → closed-loop evaluation covering understanding–generation consistency, grounded controllability, robustness, and reliability.
• Solicited topics include unified Transformer / Encoder–LLM–Decoder / autoregressive–diffusion hybrid architectures, multimodal tokenization and representation alignment, multi-task and curriculum training, data mixing, instruction tuning, preference alignment, RL and feedback optimization, synthetic data.
• Also in scope: grounded reasoning and planned/controllable editable generation, unified benchmarks and calibration, MoE, long-context video/multi-image, compression, distillation, deployment, plus applications in embodied AI, robotics, medical, remote sensing, industrial vision.
• Logistics: submissions due 1 Nov 2026 (AOE); ≥3 independent reviewers per IJCV process; conference extensions need ≥30% new contribution with disclosure; rolling review before the deadline. Special issue page: mllm-mucg.github.io/IJCV2026-SI.