Pith. sign in

REVIEW 3 cited by

PUMA: Empowering Unified MLLM with Multi-granular Visual Generation

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2410.13861 v2 pith:KAW6N2U3 submitted 2024-10-17 cs.CV

PUMA: Empowering Unified MLLM with Multi-granular Visual Generation

classification cs.CV
keywords generationmllmpumaunifiedvisualmultimodaltasksgranularity
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X LinkedIn Reddit HN
read the original abstract

Recent advancements in multimodal foundation models have yielded significant progress in vision-language understanding. Initial attempts have also explored the potential of multimodal large language models (MLLMs) for visual content generation. However, existing works have insufficiently addressed the varying granularity demands of different image generation tasks within a unified MLLM paradigm - from the diversity required in text-to-image generation to the precise controllability needed in image manipulation. In this work, we propose PUMA, emPowering Unified MLLM with Multi-grAnular visual generation. PUMA unifies multi-granular visual features as both inputs and outputs of MLLMs, elegantly addressing the different granularity requirements of various image generation tasks within a unified MLLM framework. Following multimodal pretraining and task-specific instruction tuning, PUMA demonstrates proficiency in a wide range of multimodal tasks. This work represents a significant step towards a truly unified MLLM capable of adapting to the granularity demands of various visual tasks. The code and model will be released in https://github.com/rongyaofang/PUMA.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. FLUX-Reason-6M & PRISM-Bench: A Million-Scale Text-to-Image Reasoning Dataset and Comprehensive Benchmark

    cs.CV 2025-09 conditional novelty 7.0

    The authors build a 6M-image, 20M-caption reasoning dataset with generation chain-of-thought and a 7-track VLM-judged benchmark, then rank 19 text-to-image models.

  2. One Patch Is Enough: Reinforcement-Optimized Visual Token Grounding for MLLM-Based Scene Text Spotting

    cs.CV 2026-07 conditional novelty 6.0

    Single anchor visual tokens chosen by reinforcement learning improve MLLM scene text spotting over multi-patch and coordinate-text grounding on Total-Text, CTW1500, and ICDAR 2015.

  3. One Patch Is Enough: Reinforcement-Optimized Visual Token Grounding for MLLM-Based Scene Text Spotting

    cs.CV 2026-07 conditional novelty 5.0

    Single-patch visual-token grounding with RL-based token selection and full-image geometry decoding reports large gains over multi-patch and coordinate-text grounding for MLLM scene text spotting.