Spatially Grounded Concept Bottleneck Models via Part-Factorized Attention

• URL: http://arxiv.org/abs/2606.04364v2
• Date: Wed, 10 Jun 2026 18:54:39 GMT
• Title: Spatially Grounded Concept Bottleneck Models via Part-Factorized Attention
• Abstract summary: Concept bottleneck models (CBMs) predict a layer of human-named attributes before predicting a class.<n>This work studies a part-factorized CBM that removes that freedom by construction.
• Score: 0.2578242050187029
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Concept bottleneck models (CBMs) predict a layer of human-named attributes before predicting a class, which makes their decisions auditable. On fine-grained recognition tasks the concept heads are usually free to attend anywhere in the image, so a head named for one body region can be satisfied by evidence on another. This work studies a part-factorized CBM that removes that freedom by construction. The method has three components built on a frozen DINOv3 vision transformer. A learned foreground gate, trained on DINOv3 patch features, suppresses background patches inside the part attention. A set of part queries cross-attends to patch features and each of the 312 CUB attributes is routed, through a fixed concept-to-part map, to read only from the part token its name implies. A learnable two-dimensional Gaussian prior, injected additively in log space into the attention logits, breaks the permutation symmetry among part queries; its means are initialized from the dataset-average keypoint location of each part, which requires no per-image keypoint supervision at training or test time. On CUB-200-2011 the spatial-prior model matches a fully supervised baseline (88.85% versus 88.95% top-1) while raising pointing accuracy by 16 points (52.6% versus 36.4%). Replacing bounding-box supervision with a PCA foreground target and combining it with the Gaussian prior removes all per-image supervision and reaches 88.6% top-1 at about 70% pointing accuracy. A keypoint-fraction sweep shows that 0.5% of the training set (about 27 images) suffices to initialize the prior with no measurable loss. Removing part identity entirely is the harder case: without any spatial prior, pointing accuracy collapses to $2.9\%$.

Related papers

• PatchHead: Learning Spatial Patch Evidence for Generalizable AI-Generated Image Detection [6.466210485064843]
  We introduce PatchHead, a lightweight spatial aggregation head that preserves the two-dimensional organization of DINO patch tokens and integrates evidence across neighboring regions.<n>Across nine cross-dataset benchmarks spanning manually curated and in-the-wild settings, PatchHead ranks first on seven datasets and second on the remaining two.<n>It improves the strongest prior method from 91.6% to 94.6% in average balanced accuracy (+3.0 points) and raises the worst-case accuracy from 82.4% to 89.4% (+6.9 points)
  arXiv  Detail & Related papers  (2026-08-10T07:47:22Z)
• Efficient Visual Pointing for Embodied AI:Agent-Driven Data Synthesis, Cross-Block Attention, and Iterative Correction [55.11480729304395]
  We describe our PointArena 2026 solution, which achieves 77.2% overall accuracy and ranks second on the benchmark.<n>The ap proach targets three failure modes. First, agent-driven syn thesis builds large semantic and anchor-relative candidate pools.<n>Second, a determinis tic steerable-data pipeline creates a verified 10,000-sample main set, plus reserve samples, using masks, templates, and path verification.
  arXiv  Detail & Related papers  (2026-06-29T06:39:03Z)
• Right Makes Might: Aligning Verified Hidden States Empowers RL Reasoning [55.264863369127774]
  Current methods reduce each correct rollout to a single reward bit, ignoring the geometric structure shared among their hidden states.<n>We propose Hidden-Align, an auxiliary loss function that aligns the last-layer hidden states of correct rollouts at the anchor token during RL training, with zero overhead in both training and inference.<n>On eight mathematical reasoning benchmarks, Hidden-Align improves average pass@1 over the DAPO baseline by 3.8, 6.2, and 5.4 percentage points on Qwen3-1.7B, 4B, and 14B respectively, with consistent pass@k gains across all three scales.
  arXiv  Detail & Related papers  (2026-06-02T06:51:15Z)
• Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction [51.56484100374058]
  Text detectors amplify a pretrained typicality axis; they do not construct an AI-vs-human boundary.<n>On raw encoders before any task supervision, projecting onto centroid(AI)-centroid(HC3) achieves NYT-vs-HC3 AUROC 0.806/0.944/0.834 across three architectures.<n>On RoBERTa-base, raw projection exceeds fine-tuning; on RoBERTa-base, full fine-tuning reduces discrimination below raw on both fluent-formal populations tested.
  arXiv  Detail & Related papers  (2026-05-20T19:08:38Z)
• Frozen Vision Transformers for Dense Prediction on Small Datasets: A Case Study in Arrow Localization [0.0]
  We present a system for automated detection, localization, and scoring of arrow punctures on 40,cm indoor archery target faces.<n>Our pipeline combines three components: a color-based canonical rectification stage, a frozen self-supervised vision transformer, and guided feature upsampling.<n>Across three cross-validation folds, we achieve a mean F1 score of $0.893 pm 0.011$ and a mean localization error of $1.41 pm 0.06$,mm.
  arXiv  Detail & Related papers  (2026-04-18T00:17:48Z)
• Object-Pose Estimation With Neural Population Codes [4.557963624437784]
  Object symmetry complicates the direct mapping of sensory input to object rotation.<n>We show that representing object rotation with a neural population code overcomes these limitations.<n>We achieve inference in 3.2 milliseconds on an Apple M1 CPU.
  arXiv  Detail & Related papers  (2025-02-19T03:23:43Z)
• SD-Net: Symmetric-Aware Keypoint Prediction and Domain Adaptation for 6D Pose Estimation In Bin-picking Scenarios [2.786599193929693]
  We propose a new 6D pose estimation network with symmetric-aware keypoint prediction and self-training domain adaptation (SD-Net) At the keypoint prediction stage, we designe a robust 3D keypoints selection strategy to locate 3D keypoints even in highly occluded scenes. At the domain adaptation stage, we propose the self-training framework using a student-teacher training scheme. On public Sil'eane dataset, SD-Net achieves state-of-the-art results, obtaining an average precision of 96%.
  arXiv  Detail & Related papers  (2024-03-14T12:08:44Z)
• PDiscoNet: Semantically consistent part discovery for fine-grained recognition [62.12602920807109]
  We propose PDiscoNet to discover object parts by using only image-level class labels along with priors encouraging the parts to be. Our results on CUB, CelebA, and PartImageNet show that the proposed method provides substantially better part discovery performance than previous methods.
  arXiv  Detail & Related papers  (2023-09-06T17:19:29Z)
• Centroid-centered Modeling for Efficient Vision Transformer Pre-training [44.24223088955106]
  Masked Image Modeling (MIM) is a new self-supervised vision pre-training paradigm using a Vision Transformer (ViT) Our proposed centroid-based approach, CCViT, leverages k-means clustering to obtain centroids for image modeling without supervised training of the tokenizer model. Our approach achieves competitive results with recent baselines without external supervision and distillation training from other models.
  arXiv  Detail & Related papers  (2023-03-08T15:34:57Z)
• Quality-aware Part Models for Occluded Person Re-identification [77.24920810798505]
  Occlusion poses a major challenge for person re-identification (ReID) Existing approaches typically rely on outside tools to infer visible body parts, which may be suboptimal in terms of both computational efficiency and ReID accuracy. We propose a novel method named Quality-aware Part Models (QPM) for occlusion-robust ReID.
  arXiv  Detail & Related papers  (2022-01-01T03:51:09Z)
• Completely Self-Supervised Crowd Counting via Distribution Matching [92.09218454377395]
  We propose a complete self-supervision approach to training models for dense crowd counting. The only input required to train, apart from a large set of unlabeled crowd images, is the approximate upper limit of the crowd count. Our method dwells on the idea that natural crowds follow a power law distribution, which could be leveraged to yield error signals for backpropagation.
  arXiv  Detail & Related papers  (2020-09-14T13:20:12Z)
• BRUL\`E: Barycenter-Regularized Unsupervised Landmark Extraction [2.2758845733923687]
  Unsupervised retrieval of image features is vital for many computer vision tasks where the annotation is missing or scarce. We propose a new unsupervised approach to detect the landmarks in images, validating it on the popular task of human face key-points extraction. The method is based on the idea of auto-encoding the wanted landmarks in the latent space while discarding the non-essential information.
  arXiv  Detail & Related papers  (2020-06-20T20:04:00Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.