# DINOv3

Remote sensing change detection usually depends on large collections of temporally aligned, pixel-level annotations, which limits scaling across regions and platforms. SpotDiff composes DINOv3 semantic features, AnyUp pixel-resolution feature reconstruction and SAM 3 text-guided region proposals into a training-free pipeline, then compares the cosine similarity of the two dates' reconstructed features inside the proposed regions. On LEVIR-CD the ViT-7B version reaches 82.46 F1, 2.06 points above the strongest prior training-free method; on SECOND it raises building-change IoU by 10.3.

Remote sensing change detection usually depends on large collections of temporally aligned, pixel-level annotations, which limits scaling across regions and platforms. SpotDiff composes DINOv3 semantic features, AnyUp pixel-resolution feature reconstruction and SAM 3 text-guided region proposals into a training-free pipeline, then compares the cosine similarity of the two dates' reconstructed features inside the proposed regions. On LEVIR-CD the ViT-7B version reaches 82.46 F1, 2.06 points above the strongest prior training-free method; on SECOND it raises building-change IoU by 10.3.

EEG foundation models transfer poorly as frozen feature extractors, so SeeEEG takes another route: it renders EEG as images using the montage, filtering and contrast settings of a clinical EEG viewer and trains only those display parameters, leaving the vision backbone frozen. With a frozen DINOv3 ViT-B as encoder, SeeEEG exceeds seven EEG foundation models in balanced accuracy on all eight clinical, sleep, affective and brain-computer interface datasets, reaching a mean rank of 1.3 by Cohen or AUROC against 2.9 for the strongest baseline, REVE.

EEG foundation models transfer poorly as frozen feature extractors, so SeeEEG takes another route: it renders EEG as images using the montage, filtering and contrast settings of a clinical EEG viewer and trains only those display parameters, leaving the vision backbone frozen. With a frozen DINOv3 ViT-B as encoder, SeeEEG exceeds seven EEG foundation models in balanced accuracy on all eight clinical, sleep, affective and brain-computer interface datasets, reaching a mean rank of 1.3 by Cohen or AUROC against 2.9 for the strongest baseline, REVE.

Swiss forest composition was previously shown as one dominant species per pixel, erasing the share of accompanying species. A deep-learning model anchored on 0.1 m aerial RGB imagery and combined with airborne laser scanning, annual AlphaEarth satellite embeddings and climate variables was trained on basal-area observations from 3,992 Swiss National Forest Inventory sites to output basal-area shares for ten main tree species and an Other class on a 30 m grid. Spatial cross-validation gave a macro coefficient of determination of 0.485; on forest-reserve data not used in model development, the macro value was 0.552 at 30 m pixel support and 0.733 at plot support, with dominant-class identification in 73.2% of pixels and 81.1% of plots.

Swiss forest composition was previously shown as one dominant species per pixel, erasing the share of accompanying species. A deep-learning model anchored on 0.1 m aerial RGB imagery and combined with airborne laser scanning, annual AlphaEarth satellite embeddings and climate variables was trained on basal-area observations from 3,992 Swiss National Forest Inventory sites to output basal-area shares for ten main tree species and an Other class on a 30 m grid. Spatial cross-validation gave a macro coefficient of determination of 0.485; on forest-reserve data not used in model development, the macro value was 0.552 at 30 m pixel support and 0.733 at plot support, with dominant-class identification in 73.2% of pixels and 81.1% of plots.

Patient-level ultrasound diagnosis has relied on a single attention bottleneck to aggregate multiple views, which dilutes sparse malignant cues and optimizes only global accuracy. C2LUE replaces that bottleneck with multi-query cross-attention, letting several learnable queries extract complementary morphological evidence, and adds a differentiable sensitivity-specificity geometric-mean regularizer during training. On biopsy-proven thyroid and breast ultrasound datasets, patient-level AUC reaches 0.932 and 0.936, with sensitivity rising to 0.874 and 0.876 at a moderate cost in specificity.

Patient-level ultrasound diagnosis has relied on a single attention bottleneck to aggregate multiple views, which dilutes sparse malignant cues and optimizes only global accuracy. C2LUE replaces that bottleneck with multi-query cross-attention, letting several learnable queries extract complementary morphological evidence, and adds a differentiable sensitivity-specificity geometric-mean regularizer during training. On biopsy-proven thyroid and breast ultrasound datasets, patient-level AUC reaches 0.932 and 0.936, with sensitivity rising to 0.874 and 0.876 at a moderate cost in specificity.

For the Explainable Deepfake Detection Challenge on XPlainVerse, the authors present a modular detection-and-explanation system in which a detector fuses several DINOv3 backbones with Mesorch forensic features, a Grounding-DINO pipeline converts local artifact descriptions from explanations into weak patch-level pseudo-masks supervising an Artifact Evidence Map, and a local patch-level contrastive objective separates artifact from authenticity evidence, while class-conditional Qwen3-VL models generate complex explanations and a GRPO-optimized model simplifies them, reaching 0.9349 detection accuracy, a 0.5571 explanation score, and a 0.7456 final challenge score on the full test split.

For the Explainable Deepfake Detection Challenge on XPlainVerse, the authors present a modular detection-and-explanation system in which a detector fuses several DINOv3 backbones with Mesorch forensic features, a Grounding-DINO pipeline converts local artifact descriptions from explanations into weak patch-level pseudo-masks supervising an Artifact Evidence Map, and a local patch-level contrastive objective separates artifact from authenticity evidence, while class-conditional Qwen3-VL models generate complex explanations and a GRPO-optimized model simplifies them, reaching 0.9349 detection accuracy, a 0.5571 explanation score, and a 0.7456 final challenge score on the full test split.

The authors introduce ViRA, a planner-agnostic visual representation alignment framework that supervises the planner encoder with frozen visual foundation model (VFM) intermediate features during training and discards the alignment modules at inference, leaving planner architecture and inference cost unchanged; across NAVSIM v2 and HUGSIM, ViRA consistently improves driving performance for regression-based, diffusion-based, and scoring-based planners, with gains depending on the VFM target and on whether the planner uses auxiliary perception supervision.

The authors introduce ViRA, a planner-agnostic visual representation alignment framework that supervises the planner encoder with frozen visual foundation model (VFM) intermediate features during training and discards the alignment modules at inference, leaving planner architecture and inference cost unchanged; across NAVSIM v2 and HUGSIM, ViRA consistently improves driving performance for regression-based, diffusion-based, and scoring-based planners, with gains depending on the VFM target and on whether the planner uses auxiliary perception supervision.

The work presents a training-free, one-shot framework for scientific image segmentation that uses a single annotated reference image to combine frozen DINOv3 with SAM: a background direction estimated from the reference background drives background-adaptive orthogonal projection of DINOv3 patch features to suppress artifact-related directions, cosine similarity then localizes candidate regions, and generated prompts feed SAM. Evaluated on red-blood-cell microscopy, structured-illumination pool boiling, and chest radiography, it improves mean IoU over the strongest baseline by 5.15 points on microscopy and 33.36 points on pool boiling, while matching SAM2 on chest radiographs.

The work presents a training-free, one-shot framework for scientific image segmentation that uses a single annotated reference image to combine frozen DINOv3 with SAM: a background direction estimated from the reference background drives background-adaptive orthogonal projection of DINOv3 patch features to suppress artifact-related directions, cosine similarity then localizes candidate regions, and generated prompts feed SAM. Evaluated on red-blood-cell microscopy, structured-illumination pool boiling, and chest radiography, it improves mean IoU over the strongest baseline by 5.15 points on microscopy and 33.36 points on pool boiling, while matching SAM2 on chest radiographs.

The work proposes DyRA, an input-adaptive method that dynamically approximates and corrects the output residual error introduced by structured weight approximations (such as low-rank factorizations) during inference by directly optimizing low-rank factors of the output, yielding a more faithful approximation of full matrix multiplication under the same computational budget; across vision, speech, and language models, DyRA consistently improves the accuracy-efficiency trade-off over structured weight approximations alone, and notably achieves a 1.5x end-to-end GPU speedup for DINOv3 while reducing accuracy degradation by more than 3x.

The work proposes DyRA, an input-adaptive method that dynamically approximates and corrects the output residual error introduced by structured weight approximations (such as low-rank factorizations) during inference by directly optimizing low-rank factors of the output, yielding a more faithful approximation of full matrix multiplication under the same computational budget; across vision, speech, and language models, DyRA consistently improves the accuracy-efficiency trade-off over structured weight approximations alone, and notably achieves a 1.5x end-to-end GPU speedup for DINOv3 while reducing accuracy degradation by more than 3x.