PredErase is a training‑free method that removes objects and their photometric effects by guiding a frozen Fill model with predictive latent cues. It expands the user‑provided mask to a contact‑band region, uses I‑JEPA to generate a context‑conditioned target for the hole, and aligns Fill’s completions with this target while keeping surrounding pixels fixed. On benchmarks such as RemovalBench, RORD‑Val, and DEFACTO‑Val, PredErase improves the native FLUX.2 backbone for instance‑only masks, though supervised removers still outperform it on full‑image metrics.
arXiv:2606. 28094v1 Announce Type: cross Abstract: Real-world object removal is challenging due to two key difficulties: the target object's non-local effects, such as shadows and reflections, which are difficult to model, and the fact that user-provided masks are often inaccurate or incomplete.
DART is a new RGB‑D pretraining method for surgical vision foundation models that incorporates pseudo‑labeled depth maps as a pixel‑space reconstruction target during training. By adding a depth reconstruction head to DINOv2’s masked iBOT framework, DART improves representation quality without affecting downstream RGB‑only fine‑tuning or inference. Across eight surgical benchmarks—including segmentation, depth estimation, and image‑level recognition—DART outperforms both natural‑image and in‑domain baselines, demonstrating that geometric pseudo‑labels can strengthen foundation model pretraining without extra labels or inference cost.
arXiv:2608. 02058v1 Announce Type: new Abstract: Source-free class unlearning erases whole classes using only the forget data, judged at the representation level, where features can leak a class the head no longer predicts.
arXiv:2606. 04364v1 Announce Type: cross Abstract: Concept bottleneck models (CBMs) predict a layer of human-named attributes before predicting a class, which makes their decisions auditable.
The paper introduces SGMA, a structure‑guided masked autoencoding framework designed for ultra‑high‑resolution scientific images. SGMA combines a content‑adaptive quadtree tokenizer that reduces gigapixel images to a fixed‑length sequence with a structure‑conditioned masking process that focuses reconstruction on spatially informative regions. The method, enhanced by Damped Accumulation to stabilize multi‑scale signals, achieves superior performance over standard MAE baselines on electron microscopy, whole‑slide optical microscopy, and X‑ray CT datasets, delivering significant accuracy gains and up to a 24.8× inference speedup.
arXiv:2608.28216v1 Announce Type: new Abstract: Locating a specific object instance in a cluttered scene using a single reference image and a short description, and reporting when that instance is ab...
arXiv:2607. 21318v1 Announce Type: cross Abstract: Replacing an object with one that differs in category or shape requires complete source removal, natural target formation unconstrained by the source silhouette, and preservation of unrelated content.
The paper introduces OVIE, a monocular novel-view synthesis method that eliminates the need for multi‑view training data. By using a frozen depth estimator to generate pseudo‑target views from single images and applying masked and adversarial losses, OVIE is trained on 30 million uncurated images. It achieves state‑of‑the‑art performance on RealEstate10K and DL3DV, produces highly consistent multi‑view trajectories, and runs at 116 FPS—over 600× faster than the fastest baseline.
arXiv:2607.05568v2 Announce Type: replace-cross Abstract: Compact primitive abstractions represent 3D shapes with a few geometric primitives while preserving recognizable components. Learned methods...
arXiv:2608. 05774v1 Announce Type: cross Abstract: Joint-embedding predictive architectures learn by predicting latent representations of missing observations, yet many masked JEPAs are evaluated primarily through the encoders they produce.
LeVJEPA is a video encoder that eliminates the need for architectural asymmetries, exponential-moving-average target encoders, stop-gradients, and capacity-limited predictors used in prior self‑supervised methods. It trains a single encoder with an invariance loss over global and local views, regularized by SIGReg to prevent collapse, and achieves strong performance with far less pretraining compute. The approach also allows block‑causal attention, making temporal ordering a property of the encoder itself, and matches or surpasses state‑of‑the‑art baselines on both appearance‑centric and motion‑centric benchmarks.
Compositional analysis of frozen vision encoders should determine both what changed and where it changed. Standard factor probes score these axes separately, however, and can reward multiple operations that reuse the same predicted slot.