The paper introduces D2MoE, a Mixture of Task Adapters designed for vision foundation models to handle multi-task dense prediction. It separates adaptation into a task‑generic low‑rank channel (TG‑LoRA) that preserves pretrained priors and a task‑adaptive sparse expert module (TA‑MoE) that resolves task‑specific conflicts. A Semantic Prior Injector further aligns multi‑level task features with frozen foundation knowledge, achieving robust multi‑task performance with a small trainable footprint.
arXiv:2610.01942v1 Announce Type: new Abstract: Predicting the future evolution of a scene is a fundamental capability for world modeling. Recent work has shown that operating in the feature space of...
The paper introduces MemMTL, a multi‑task dense prediction framework that uses a compact task state derived from global visual context and refines it via a learnable prototype memory. This refined state informs task‑conditioned expert logits, which are combined with token‑level logits and routed through a sparse top‑k selection over a shared local expert bank. A task‑agnostic residual bank offers a common adaptation path, and both paths are added to the backbone feature before task‑specific prediction. The authors outline an evaluation protocol on NYUD‑v2 and PASCAL‑Context using SAM 3 and ViT‑L backbones to assess predictive quality, computational cost, and the contributions of task‑state conditioning, prototype retrieval, and sparse routing.
arXiv:2610.23533v1 Announce Type: new Abstract: Recent image generators exhibit rich visual priors that can support dense visual perception. However, existing approaches typically encode multiple per...
arXiv:2605.12491v2 Announce Type: replace Abstract: Vision Transformers (ViTs) learn rich visual-semantic representations through all-to-all self-attention among patch tokens. However, this design im...
arXiv:2607. 26107v1 Announce Type: cross Abstract: Dense vision-language understanding, including object localization, region recognition, and open-vocabulary semantic segmentation, requires associating language concepts with spatially grounded visual regions.
The paper introduces Lens, a training‑free framework that aligns multimodal representations with the semantic perspective required by downstream tasks. Lens uses a task‑specific readout phrase to anchor the perspective and then aggregates token states after the full input, ensuring the extracted representation reflects task‑conditioned evidence integration rather than generic salient content. The method achieves a Precision@1 of 63.9 across 36 MMEB datasets, outperforming the nearest training‑free baseline by 10.2 points.
arXiv:2609.21522v1 Announce Type: new Abstract: Recent pre-trained foundation models provide rich multi-modal priors for downstream 3D vision tasks. However, the effectiveness of these representation...
arXiv:2603. 22282v2 Announce Type: replace-cross Abstract: We present UniMotion, to our knowledge the first unified framework for simultaneous understanding and generation of human motion, natural language, and RGB images within a single architecture.
VIVAS is a new Vision‑Language Model pre‑training framework that addresses the lack of fine‑grained visual perception in existing VLMs. It introduces a unified token space and a dense‑structural‑semantic vision tokenizer that expands the textual vocabulary with visual tokens, enabling vision‑language unified autoregressive supervision over both visual details and linguistic content. Trained on 12.4 T tokens, VIVAS achieves state‑of‑the‑art results on 7 tasks and 39 multimodal benchmarks.
arXiv:2604. 07753v2 Announce Type: replace-cross Abstract: Empowering Large Multimodal Models (LMMs) with image generation often leads to catastrophic forgetting in understanding tasks due to severe gradient conflicts.
arXiv:2607.01630v2 Announce Type: replace Abstract: Dynamic expansion methods for class-incremental learning (CIL) protect task-specific knowledge by growing dedicated tokens or subnetworks, yet our...
DiscoVL introduces a disentangled cross‑modal representation learning framework for vision‑language models, combining orthogonal adversarial regularization with a multi‑branch low‑rank residual aligner to enhance bidirectional visual‑textual interaction. The method decomposes representations into subspaces, enabling cross‑modal feedback at each layer and preventing centroid collapse through an orthogonal adversarial triplet loss. Across 15 benchmarks, DiscoVL consistently outperforms state‑of‑the‑art approaches in base‑to‑novel generalization, cross‑dataset evaluation, and few‑shot learning.