# Queries Knew More Than We Thought: Uncovering Latent Knowledge in Segmentation Models

arXiv: https://arxiv.org/abs/2609.20283 · HTML: https://mlsanity.com/paper/2609.20283/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 36 of its 50 references that are in the map.
- First posted on arXiv 30 Jul 2026.

## Builds on (most cited first, 30 of 36)

- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [End-to-End Object Detection with Transformers](https://mlsanity.com/paper/2005.12872/) (2020)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [The Cityscapes Dataset for Semantic Urban Scene Understanding](https://mlsanity.com/paper/1604.01685/) (2016)
- [Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection](https://mlsanity.com/paper/2303.05499/) (2023)
- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [Masked-attention Mask Transformer for Universal Image Segmentation](https://mlsanity.com/paper/2112.01527/) (2021)
- [Visual Prompt Tuning](https://mlsanity.com/paper/2203.12119/) (2022)
- [DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection](https://mlsanity.com/paper/2203.03605/) (2022)
- [Semantic Understanding of Scenes through the ADE20K Dataset](https://mlsanity.com/paper/1608.05442/) (2016)
- [SAM 3: Segment Anything with Concepts](https://mlsanity.com/paper/2511.16719/) (2025)
- [COCO-Stuff: Thing and Stuff Classes in Context](https://mlsanity.com/paper/1612.03716/) (2016)
- [AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition](https://mlsanity.com/paper/2205.13535/) (2022)
- [Mixture-of-Experts with Expert Choice Routing](https://mlsanity.com/paper/2202.09368/) (2022)
- [DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR](https://mlsanity.com/paper/2201.12329/) (2022)
- [Continual Test-Time Domain Adaptation](https://mlsanity.com/paper/2203.13591/) (2022)
- [DN-DETR: Accelerate DETR Training by Introducing Query DeNoising](https://mlsanity.com/paper/2203.01305/) (2022)
- [Soft-NMS -- Improving Object Detection With One Line of Code](https://mlsanity.com/paper/1704.04503/) (2017)
- [Segment Anything in High Quality](https://mlsanity.com/paper/2306.01567/) (2023)
- [Mask DINO: Towards A Unified Transformer-based Framework for Object Detection and Segmentation](https://mlsanity.com/paper/2206.02777/) (2022)
- [OneFormer: One Transformer to Rule Universal Image Segmentation](https://mlsanity.com/paper/2211.06220/) (2022)
- [Scaling & Shifting Your Features: A New Baseline for Efficient Model Tuning](https://mlsanity.com/paper/2210.08823/) (2022)
- [Generalized Decoding for Pixel, Image, and Language](https://mlsanity.com/paper/2212.11270/) (2022)
- [Mask Scoring R-CNN](https://mlsanity.com/paper/1903.00241/) (2019)
- [LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation](https://mlsanity.com/paper/2110.08733/) (2021)
- [Personalize Segment Anything Model with One Shot](https://mlsanity.com/paper/2305.03048/) (2023)
- [From Sparse to Soft Mixtures of Experts](https://mlsanity.com/paper/2308.00951/) (2023)
- [Matcher: Segment Anything with One Shot Using All-Purpose Feature Matching](https://mlsanity.com/paper/2305.13310/) (2023)
- [Large-scale Unsupervised Semantic Segmentation](https://mlsanity.com/paper/2106.03149/) (2021)

## Abstract (arXiv)

Modern segmenters often fail after the expensive computation has already been done: a useful mask is present among the model's query-conditioned candidates, but the deployed selection rule does not expose it. We study this output-selection bottleneck in frozen DETR-family models. A ground-truth-only oracle first shows substantial hidden headroom in already-computed mask proposals. This raises a simple question: How can we better use the masks a segmenter has already computed but does not expose? We then ask whether that headroom can be recovered without adding queries, generating new masks, rerunning the backbone, or updating weights. HYDRA is a small selector trained only on cached frozen outputs. At inference time, it scores the cached candidates against an explicit keep-baseline option and acts only when a held-out calibrated margin indicates the selected candidate is sufficiently better. Trained on training-split caches and calibrated on held-out data, HYDRA improves Mask2Former, MaskDINO, and OneFormer by up to +7.41 dataset mIoU points on ADE20k and COCO, and improves SAM 3 by +9.4 class-macro prompt-IoU points on average across eight domains while preserving useful predictions through calibration. Paired LoRA controls show that lightweight weight adaptation does not remove the bottleneck: exposed predictions are often flat or worse, while routing over the adapted candidates still recovers accuracy. Finally, we connect the effect to query specialization under bipartite matching and verify it in a controlled TinyDETR study. These results show that frozen segmenters should be evaluated not only by the masks they expose, but also by the useful candidates they suppress.