Robota-Embed tackles the problem of retrieving relevant physical experience for robots by learning a single embedding space that jointly encodes vision, language, and action from both the current state and a future trajectory segment. The core insight is that a useful retrieval key must capture not just task intent and visual scene, but also the robot's ongoing motion and execution stage, because the goal is to retrieve an appropriate continuation rather than simply a similar observation. A new multi-source benchmark and two downstream applications—in-context policy conditioning and memory evolution around a frozen large policy—show that this unified, action-aware representation significantly outperforms vision-language-only baselines and improves both success rates and token efficiency.

End-to-end accuracy on multi-image vision-language tasks hides whether failures stem from visual retrieval, schema understanding, or downstream composition. This paper introduces a within-item $2 \times 2$ factorial Oracle design in which a model performs a 16-image schema-application task under four matched conditions, receiving verified textual schema (Oracle-A), verified ordered target appearances (Oracle-B), both, or neither. The results localize the dominant recoverable deficit to upstream ordered instance recovery—Oracle-B rescues strict accuracy by 54.6 percentage points—while supplying the schema alone barely helps, suggesting that current VLMs often lose the relevant visual facts before any codebook reasoning begins.

Industrial anomaly detection is bottlenecked by scarce anomalous samples, which limits supervised training and defect understanding. This paper proposes CORE, a training-free framework that leverages Large Multimodal Models to synthesize realistic anomalies in both zero-shot and one-shot regimes through structured prompt optimization. The key idea is to decompose prompts into BASE, VARIANT, and NEGATIVE components and refine them via a self-supervised contrast-and-reflect mechanism without human annotations or task-specific fine-tuning. If effective, this could remove the need for costly real defect collections and enable open-world adaptation.

Real-time monocular human pose estimation breaks down when the lower body is occluded or the subject steps out of frame. This paper introduces a part-decoupled temporal fusion framework that augments monocular video with sparse plantar-pressure and foot-mounted acceleration from smart insoles. The core insight is that cheap, wearable foot sensors provide complementary physical constraints exactly where monocular vision is weakest, enabling robust pose and gait recovery under severe visual degradation.

LED-BENCH targets a realistic but under-evaluated setting: answering engineering questions from complete, large-format drawings in mechanical, architectural, and transportation domains. The benchmark contains 107 sheets and 377 expert-verified questions that require models to localize fine-grained evidence and integrate information across distant regions or views. Its central finding is that current multimodal models remain far from solving this problem—the best evaluated model achieves only 52.65% accuracy, and performance consistently degrades on harder reasoning tasks.

ActICL attacks the problem of suboptimal multimodal in-context learning by training an MLLM agent to iteratively acquire, inspect, and curate demonstrations through tool-augmented reasoning. Its core idea replaces static similarity retrieval with an agentic loop of search, read, curate, and solve actions, learned via a two-stage post-training pipeline of rejection-sampling SFT and joint reinforcement learning. This matters because it shows that even a modestly sized trained agent can construct contexts that substantially improve its own accuracy and also boost larger frontier answering models.

Bone suppression in chest X-rays removes overlapping ribs and clavicles that hide lung pathology, but existing methods suffer from reconstruction-biased latent compression, domain gaps, and very small paired datasets. DeBoneDiT tackles this by combining a depth-supervised auto-encoder with a source-conditioned Brownian bridge diffusion Transformer, and it releases SZCH-X-Rays, a 741-pair high-resolution dataset. The goal is to provide an efficient, software-only alternative to hardware-dependent Dual-Energy Subtraction.

Unsupervised feature selection often forces practitioners to hand-tune regularization parameters in sparse PCA models or to collect large labeled datasets for deep learning. This paper proposes MDUFS-Net and MDUFS-Net+, which unfold an ADMM solver for $\ell_{2,0}$-norm regularized PCA into a trainable deep network, embed a Stiefel manifold projection to preserve orthogonality, and optionally add element-wise $\ell_0$ sparsity. The work matters because it tries to retain the interpretability of model-driven optimization while automating parameter adaptation via gradient-based learning.

FrameEvidence is a diagnostic benchmark for multi-image spatial reasoning that tests whether vision-language models correctly answer questions when the reference frame or the set of visual evidence is manipulated. It introduces 960 Chinese multiple-choice questions arranged in 320 matched triplets, where FQ1 varies the spatial reference frame while holding images fixed and FQ2 deletes or adds images to test evidence necessity and invariance. The core finding is a sharp divergence between single-item accuracy, which reaches 50\% for GPT-5.6, and joint correctness across matched variants, which remains below 11\% for $\mathrm{FQ1\text{-}GEM}$. This design exposes how standard accuracy obscures brittle reasoning and answer instability.

Text-to-image safety alignment typically relies on coarse, image-level preference signals that can degrade overall generation quality. SaTPO proposes step-wise transition preferences between an unsafe prompt and its safe rewrite, sharing the parent latent state and noise to isolate the safety direction without a learned reward model. Multi-scale drift regularization and benign replay attempt to preserve ordinary generation, and experiments on Stable Diffusion v1.5 and SDXL report lower unsafe-detection rates and better FID than strong baselines.

This paper proposes FLARE, a framework for learning robust visual speech representations for lipreading by distilling knowledge from Whisper audio models into discrete codes via random-projection quantization, while jointly processing full-face and lip-ROI video streams. The core idea is that richer acoustic supervision and complementary facial dynamics can mitigate the ambiguity and pose sensitivity that plague visual speech recognition. FLARE achieves competitive word error rates on LRS2, LRS3, and the challenging out-of-domain WildVSR benchmark, and the authors back their design with extensive ablations and representation visualizations.

Text-to-image models increasingly generate Chinese text that is readable yet structurally defective at the stroke level, but OCR and vision-language systems miss these errors because they are optimized for recognition invariance rather than glyph fidelity. TextVerifier reframes the problem as target-conditioned anomaly detection: given a target character and an image crop, it predicts whether the crop is a correct rendering. The core idea is to replace single-character text tokens with Ideographic Description Sequences in a compact CLIP-style dual tower, training the verifier through a three-stage curriculum that culminates in label-free hard negatives mined from an inpainting model's training trajectory. The work matters because Chinese rendering errors are often invisible to string-level metrics yet fatal in publishing and education, and existing benchmarks score only whole-image OCR accuracy rather than per-character correctness.

Industrial board inspection must remain reliable across unseen lighting, surface contamination, and production-line conditions, yet training data is typically drawn from limited source scenarios. This paper treats the IDA 2026 Track 1 benchmark as a domain generalization problem and argues that standard mAP-driven checkpoint selection misaligns with the screening-heavy deployment objective. The authors propose a single-model YOLOv11-based framework with four modules—appearance-robust training (MixRand and IBN-a), small-defect preservation (SPD, SimAM, and NWD), confidence calibration (VFL and threshold search), and metric-aligned checkpoint averaging (a custom Track 1 proxy plus SWAD)—to improve cross-scenario defect detection and clean-board screening without ensembles or external data.

3D Gaussian Splatting enables high-quality real-time rendering but its millions of primitives incur heavy storage costs. RDCompGS targets compression pipelines that map Gaussian attributes into 2D grids for mature image codecs like H.265/HEVC and JPEG XL. The core idea is to close the loop between Gaussian optimization and subsequent coding by adding a differentiable coding-rate estimate, derived from DCT coefficient statistics, to the standard rendering loss, alongside an adaptive constraint that balances rate and distortion across quantization steps. By jointly optimizing for both reconstruction fidelity and compressibility during training, the method delivers substantial bit-rate savings without sacrificing rendering quality.

3D point cloud segmentation models like Point Transformer V3 use space-filling curves (SFC) to approximate nearest neighbors efficiently, but SFC adjacency can misalign with true spatial proximity. This paper proposes DAS Loss, a training-time contrastive objective that penalizes or rewards feature similarity for SFC-adjacent pairs that are spatially distant, depending on whether they share a class label. The method boosts mIoU on PTv3 without adding parameters or inference cost, offering a lightweight plug-in for serialized point cloud transformers.

This paper tackles the problem of recursive self-improvement for agentic video-understanding harnesses, asking how automated evolution can discover generalizable designs when it must learn from limited, noisy, and repeatedly reused validation feedback. The authors first diagnose three interconnected failure modes in a baseline system (validation overfitting, search-budget dilution, and the early-experience trap), then propose VideoHarness-SRSI, an evidence-governed framework that hides validation content from the proposer, warms up with controlled interventions, and evolves mechanisms in isolated probe–race branches before factorial composition. Across four benchmarks, the method improves the mean held-out score by 6.2% relative to the baseline and cuts the validation–test gap by 37.9%, supporting the central claim that how search is informed and organized matters as much as how well it optimizes validation metrics.

Wedge addresses the inefficiency of generic environment generation for embodied navigation: pretrained agents fail for different environmental reasons, so simply adding more diverse or realistic training scenes may not address their actual shortcomings. The paper proposes a policy-conditioned generation framework that diagnoses an agent's specific weaknesses by contrasting its successful and failed rollouts, uses an LLM to propose executable environmental features (such as target visibility or blind-path ratio), validates them via success-rate gaps, and generates targeted training environments. This approach is tested across three heterogeneous agents—an RL policy, an imitation-learned model, and a black-box VLM—showing consistent gains in ProcTHOR and transfer to HM3D.

MLLM-based detectors fail in dense scenes when autoregressive coordinate generation collapses into box-repetition loops, wasting the finite decoding budget on redundant predictions. This work attributes the failure to historical-answer attention takeover—raw coordinate history crowds out the visual context needed to discover new objects. The proposed Compressed Window Auto-Regression (WinAR) keeps exact autoregression inside local windows and compresses completed windows into latent tokens via a learned Window History Compressor. On the RSDense benchmark with 200–2,000 objects per image, WinAR improves RexOmni F1@IoU$=0.5$ from 32.16 to 39.36 while cutting redundancy from 35.48% to 13.00%, and it generalizes to parallel decoding and polygon outputs.

When text and images conflict in vision-language models, prior work has characterized behavior as modality-specific biases such as "blind faith in text." This paper proposes a predictive alternative: a two-parameter Bayesian cue combination rule in which each modality contributes conditionally independent evidence weighted by a constant integration bias, plus a prior. The authors show that this simple linear model accurately predicts how VLM preferences shift across dozens of input degradation conditions and even transfers to entirely new corruption types without refitting, offering a unified account of why text sometimes dominates and vision sometimes wins.

The paper solves zero-shot stitching of autoencoders: enabling an encoder from one independently trained model to work with a decoder from another. The core idea is a training-time topological regularizer based on 0-dimensional persistent homology (MST edges) that preserves the input data's geometric skeleton in each latent space, so that after training only a lightweight affine map is needed. This matters for distributed and privacy-sensitive settings where encoders and decoders cannot be jointly optimized or share weights.