{"total":12,"items":[{"citing_arxiv_id":"2605.14110","ref_index":13,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"SToRe3D: Sparse Token Relevance in ViTs for Efficient Multi-View 3D Object Detection","primary_cat":"cs.CV","submitted_at":"2026-05-13T20:53:03+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"SToRe3D delivers up to 3x faster inference for multi-view 3D object detection in ViTs by selecting relevant 2D tokens and 3D queries via mutual relevance heads with only marginal accuracy loss.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12430","ref_index":18,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"AOI-SSL: Self-Supervised Framework for Efficient Segmentation of Wire-bonded Semiconductors In Optical Inspection","primary_cat":"cs.CV","submitted_at":"2026-05-12T17:27:25+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"AOI-SSL combines small-domain self-supervised pre-training of vision transformers with in-context patch retrieval to reduce labeled data needs and enable fast adaptation for semiconductor wire-bond segmentation.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"train-test split, with an additional set of∼100 labeled images for validation. In single-device retrieval experiments, the memory consists of 42 curated images, evaluated on a dis- joint set of 24 images from the same device family to ensure that structural priors are not trivialized by image repetition. Baseline Selection and Tuning.We select ResNet18 [ 18], U-Net++ [ 41] and DeepLabV3+ [ 9] as primary convolutional baselines. These architectures approximate industry standards for high-precision defect detection and high-throughput industrial segmentation, respectively. To ensure a fair comparison, ImageNet- pretrained baselines are trained with an inverted layer-wise learning rate decay; higher learning rates are set in earlier"},{"citing_arxiv_id":"2605.10349","ref_index":11,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Portable Active Learning for Object Detection","primary_cat":"cs.CV","submitted_at":"2026-05-11T10:52:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"PAL is a portable active learning method for object detection that uses class-specific logistic classifiers for uncertainty and image-level diversity to select annotation batches, showing better label efficiency than baselines on COCO, VOC, and BDD100K.","context_count":1,"top_context_role":"method","top_context_polarity":"use_method","context_text":"Our implementation was built on PPAL's [27] implemen- tation of MMDetection toolkit [5]. For RetinaNet experi- ments the training schedule spanned 26 epochs across all three datasets, with a learning rate decay factor of 0.1 ap- plied at the20 th epoch, following [27] code. Faster R-CNN hyperparameters followed DivProto [26], while SSD used the settings from MIAL [30]. ResNet-50 [11] served as the default backbone architecture for RetinaNet and Faster RCNN while VGG-16 [22] was used as the base detector for SSD, consistent with [30]. While most prior active learning works evaluate large detectors, practical deployments e.g., automotive, com- monly rely on edge-friendly models training on large-scale datasets. To assess PAL on such edge models, we experi-"},{"citing_arxiv_id":"2605.08320","ref_index":29,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Improved monocular depth prediction using distance transform over pre-semantic contours with self-supervised neural networks","primary_cat":"eess.IV","submitted_at":"2026-05-08T16:20:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Self-supervised monocular depth estimation improves in low-texture regions by using distance transforms on jointly estimated pre-semantic contours to create more informative loss signals.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"evaluation, following [16, 44], with test images resized to 320×256pixels. 5.2. Implementation Details We emphasize that our method is compatible with any DNN architecture and input image size. For fair comparison, our experiments are done with a256×832resolution and a UNet [62] structure, following [91], unless stated other- wise. The depth and edge networks share a ResNet [29] encoder. Our baseline, CoopNet [28], handles moving ob- jects-especially useful in Cityscapes-via self-supervised flow. The pose network, based on a ResNet encoder, outputs a 6-DoF vector. All encoders use ResNet-50 backbones ini- tialized with ImageNet [63] weights. Training runs for 30 epochs with batch size 4, initial learning rate of10 −4 (de-"},{"citing_arxiv_id":"2605.07188","ref_index":14,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"PicoEyes: Unified Gaze Estimation Framework for Mixed Reality with a Large-Scale Multi-View Dataset","primary_cat":"cs.CV","submitted_at":"2026-05-08T03:34:09+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"PicoEyes delivers a unified end-to-end model for full 3D gaze estimation including eye parameters, axes, segmentation and depth from monocular or binocular near-eye images, supported by a new large-scale multi-view dataset.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.05590","ref_index":25,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Uncertainty-Guided Edge Learning for Deep Image Regression in Remote Sensing","primary_cat":"cs.CV","submitted_at":"2026-05-07T02:17:17+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"UGEL employs deep beta regression to estimate uncertainty in one forward pass, enabling faster convergence in edge learning for remote sensing image regression than active or semi-supervised baselines.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.24119","ref_index":9,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"TopoHR: Hierarchical Centerline Representation for Cyclic Topology Reasoning in Driving Scenes with Point-to-Instance Relations","primary_cat":"cs.CV","submitted_at":"2026-04-27T07:13:33+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"TopoHR introduces hierarchical point/instance/semantic queries and a unified P2I+I2I topology module that reports SOTA gains on OpenLane-V2 subsets.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.15166","ref_index":17,"ref_count":2,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Class Unlearning via Depth-Aware Removal of Forget-Specific Directions","primary_cat":"cs.CV","submitted_at":"2026-04-16T15:46:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"DAMP performs one-shot class unlearning by depth-aware projection removal of forget-specific directions, producing forgetting behavior closer to retraining from scratch than prior methods on image classification tasks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.07122","ref_index":10,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Accuracy Improvement of Semi-Supervised Segmentation Using Supervised ClassMix and Sup-Unsup Feature Discriminator","primary_cat":"cs.CV","submitted_at":"2026-04-08T14:18:00+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Supervised ClassMix and a Sup-Unsup Feature Discriminator yield an average 2.07% mIoU gain over standard semi-supervised methods on Chase and COVID-19 datasets.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.06795","ref_index":6,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"FedDAP: Domain-Aware Prototype Learning for Federated Learning under Domain Shift","primary_cat":"cs.CV","submitted_at":"2026-04-08T08:08:33+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"FedDAP improves federated learning under domain shift by creating domain-specific global prototypes via similarity-weighted fusion and using them for domain-aware local feature alignment.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2512.10421","ref_index":6,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"Neural Collapse in Test-Time Adaptation","primary_cat":"cs.CV","submitted_at":"2025-12-11T08:34:58+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Sample-wise neural collapse reveals that feature-classifier misalignment drives TTA degradation under shifts, which NCTTA corrects via hybrid geometric-predictive targets.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2511.16428","ref_index":15,"ref_count":1,"confidence":0.55,"is_internal_anchor":false,"paper_title":"CylinderDepth: Cylindrical Spatial Attention for Multi-View Consistent Self-Supervised Surround Depth Estimation","primary_cat":"cs.CV","submitted_at":"2025-11-20T14:55:28+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"CylinderDepth uses cylindrical spatial attention with non-learned weights to enforce cross-view consistency in self-supervised surround depth estimation.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}