Skip to content
Computational Pathology Reading Roadmap

Detection and segmentation

Dense prediction asks not only what is present but where, at what scale, and with which boundary or identity. Detection, semantic segmentation, instance segmentation, panoptic segmentation, phrase grounding, and promptable masks are related but non-interchangeable tasks.

Latest literature discovery audit: 2 October 2026. Existing paper reviews retain their individual source-check dates.

ECCV 2026 conference additions and metadata checked through 2026-09-12; the broader literature audit boundary above is unchanged.

New work belongs under proposal, query, mask, grounding, or prompt mechanisms—not under a dated update heading.

Prerequisites and objectives

Section titled “Prerequisites and objectives”

Complete image foundations and recognition. You should understand convolutional feature maps, transfer learning, cross-entropy, IoU, and ranked predictions. This chapter should enable you to:

  • derive targets and losses for boxes, classes, masks, and text-grounded regions;
  • distinguish two-stage, one-stage, and set-prediction detectors;
  • compare semantic, instance, and panoptic outputs;
  • explain what language supervision changes in open-vocabulary perception;
  • evaluate localization, classification, boundary, object, and prompt errors separately.

Task definitions

Section titled “Task definitions”

For ground-truth box B and prediction B̂,

IoU(B, B̂) = |B ∩ B̂| / |B ∪ B̂|

Detection produces a ranked set {(B̂_i, ŷ_i, ŝ_i)} for i = 1…N. Average precision therefore depends on confidence order, class, matching threshold, duplicate handling, and maximum detections. A single “mAP” without the protocol is incomplete.

Segmentation outputs differ:

  • semantic: one category per pixel, with no instance identity;
  • instance: a ranked mask per countable object;
  • panoptic: one non-overlapping assignment covering “things” and “stuff”;
  • promptable: a mask conditional on a point, box, mask, text, or other prompt;
  • open-vocabulary: recognition or localization for concepts beyond a fixed training label set.

Open-set, open-vocabulary, zero-shot, and open-world are not synonyms. State which classes, images, annotations, and textual concepts were absent during training.

Region proposals and one-stage detection

Section titled “Region proposals and one-stage detection”

Two-stage systems first propose object-like regions and then classify/refine them. One-stage systems predict densely and must manage foreground/background imbalance.

Stage Primary reading · year · venue/status Read it for
D1 R-CNN — 2014 · CVPR, peer reviewed region proposals plus transferred CNN features
D2 Fast R-CNN — 2015 · ICCV, peer reviewed shared feature computation and region-of-interest pooling
D3 Faster R-CNN — 2015 · NeurIPS, peer reviewed a learned region-proposal network and coupled proposal/detection training
D4 You Only Look Once — 2016 · CVPR, peer reviewed detection as one dense prediction problem and the accuracy–speed trade-off
D5 Feature Pyramid Networks — 2017 · CVPR, peer reviewed semantically strong features at several spatial scales
D6 Focal Loss for Dense Object Detection — 2017 · ICCV, peer reviewed foreground imbalance and loss reweighting in RetinaNet
D7 FCOS — 2019 · ICCV, peer reviewed anchor-free per-pixel box prediction, feature-level assignment, and center-ness suppression of low-quality boxes

Anchors, matching rules, box transforms, NMS thresholds, image scale, and maximum predictions are model components. Record them rather than describing only the backbone.

Detection as set prediction

Section titled “Detection as set prediction”

DETR predicts a fixed-size set of objects. If σ is the optimal bipartite assignment between targets and predictions, training minimizes a matched sum of classification and localization costs:

L_set = min_σ Σ_i [
  L_cls(y_i, ŷ_σ(i))
  + λ₁ ‖b_i − b̂_σ(i)‖₁
  + λ_IoU L_GIoU
]

This removes hand-designed anchor matching and NMS from the core formulation, but introduces learned queries, assignment costs, decoder depth, and convergence issues.

Stage Primary reading · year · venue/status Mechanism to isolate
Q1 DETR — 2020 · ECCV, peer reviewed object queries, global matching, and end-to-end set loss
Q2 Deformable DETR — 2021 · ICLR, peer reviewed sparse multiscale sampling and faster convergence
Q3 DINO: DETR with Improved DeNoising Anchor Boxes — 2023 · ICLR, peer reviewed denoising training, mixed query selection, and iterative refinement
Q4 RT-DETR — 2024 · CVPR, peer reviewed efficient hybrid encoding, uncertainty-minimal query selection, and NMS-free real-time set detection
Q5 CellDETR — 2026 · arXiv preprint specializes set detection to dense histology cells and reuses detection as a scalable representation-learning signal; domain evidence remains preprint-level

This DINO is a detector. It is distinct from DINO self-distillation in the representation chapter.

Distilling dense representations

Section titled “Distilling dense representations”
Primary reading · year · venue/status Dense-prediction transfer question
Distilling Knowledge via Knowledge Review — 2021 · CVPR, peer reviewed whether cross-stage feature review transfers across detection and segmentation heads under matched backbones
Decoupled Knowledge Distillation — 2022 · CVPR, peer reviewed how target/non-target logit decomposition interacts with dense foreground/background imbalance
Masked Generative Distillation — 2022 · ECCV, peer reviewed whether reconstructing masked teacher features transfers useful spatial context without forcing every feature value

Use identical detector/segmenter heads, assignment rules, image scales, augmentation, schedule, and inference post-processing. Otherwise the experiment mixes transfer-interface effects with a different dense-prediction recipe.

From pixels to universal masks

Section titled “From pixels to universal masks”

Encoder–decoder systems restore spatial resolution with upsampling, skip connections, pyramids, or dilation. Query-based systems predict mask embeddings and associate them with pixel features.

Stage Primary reading · year · venue/status Read it for
S1 Fully Convolutional Networks — 2015 · CVPR, peer reviewed converting classification features into dense scores
S2 U-Net — 2015 · MICCAI, peer reviewed symmetric decoding and high-resolution skip features
S3 Pyramid Scene Parsing Network — 2017 · CVPR, peer reviewed multiscale context for scene parsing
S4 Mask R-CNN — 2017 · ICCV, peer reviewed aligned region features and a parallel instance-mask head
S5 DeepLabv3+ — 2018 · ECCV, peer reviewed atrous context plus boundary-refining decoder
S6 Panoptic Feature Pyramid Networks — 2019 · CVPR, peer reviewed joining instance “things” and semantic “stuff”
S7 Mask2Former — 2022 · CVPR, peer reviewed masked cross-attention and one architecture for semantic, instance, and panoptic tasks

Pixel overlap can hide object errors. A large merged foreground may have acceptable Dice while failing to count, separate, or localize small instances. Include boundary and object metrics when those errors matter.

Language-grounded and promptable perception

Section titled “Language-grounded and promptable perception”

Global image–text alignment makes class names usable as classifiers, but a global embedding does not automatically localize a phrase. Region supervision, cross-modal fusion, or a separate proposal/mask mechanism is still needed.

Branch Primary reading · year · venue/status Valid claim boundary
global alignment CLIP — 2021 · ICML, peer reviewed; SigLIP — 2023 · ICCV, peer reviewed image–text retrieval and zero-shot scoring, not pixel grounding
grounded detection GLIP — 2022 · CVPR, peer reviewed reformulating detection as phrase grounding with cross-modal pretraining
open-vocabulary panoptic ODISE — 2023 · CVPR, peer reviewed transferring frozen text-to-image diffusion features to category-aware panoptic masks
open-set queries Grounding DINO — 2024 · ECCV, peer reviewed language-conditioned boxes; disclose the paper’s operational definition of zero-shot
promptable images Segment Anything — 2023 · ICCV, peer reviewed class-agnostic masks from spatial prompts and a data-engine loop
promptable video SAM 2 — 2025 · ICLR, peer reviewed streaming memory for prompted image/video masks
retrieval-style open vocabulary WeDetect — 2026 · CVPR, peer reviewed a dual-tower detection/retrieval design; test vocabulary, speed, and rare concepts separately

SAM and SAM 2 do not assign semantic class names by themselves. Combining a grounded detector with a promptable mask model is a pipeline; report the detector and mask contributions separately.

Controlled lab

Section titled “Controlled lab”

Choose a small detection/segmentation dataset with redistributable annotations.

  1. Freeze one backbone and input resolution.
  2. Compare a proposal-based detector with a query-based detector, or compare a U-Net/FPN-style decoder with a mask-query decoder.
  3. Keep train/validation/test images and annotation conversions identical.
  4. Report COCO AP with IoU and size breakdowns for detection; class mIoU, Dice, boundary score, object-level F1, or PQ as appropriate for segmentation.
  5. Inspect false positives, duplicate detections, localization errors, missed small objects, merged/split instances, boundary errors, and ignore-label behavior.
  6. For an open-vocabulary extension, pre-register base and novel concepts, prompt templates, synonyms, background terms, and whether evaluation images or labels entered pretraining.
  7. Measure preprocessing plus inference latency, peak memory, and maximum prediction count on named hardware.

Failure modes

Section titled “Failure modes”
  • object instances cross the split because video frames or source scenes were split independently;
  • a detector’s AP changes through NMS, test resolution, or maximum detections rather than learned representation;
  • resizing or annotation conversion moves small boxes and boundaries;
  • background and “void” labels are handled differently between methods;
  • a text prompt contains attributes absent from the image, but category-name AP hides the grounding error;
  • vocabulary overlap is described as visual generalization without auditing pretraining;
  • a promptable mask is presented as autonomous semantic segmentation;
  • mean IoU hides disappearance of rare classes or small objects.

Self-check and exit task

Section titled “Self-check and exit task”
  1. Why does AP depend on ranking as well as correctness?
  2. Which roles are played by proposals, RoIAlign, FPN, NMS, and focal loss?
  3. What replaces anchor matching and NMS in the DETR formulation?
  4. When can Dice improve while instance quality becomes worse?
  5. What additional evidence is needed to call a result open-vocabulary?
  6. Why is a Grounding-DINO-plus-SAM pipeline not one grounded segmentation model?

Exit task: write an error ledger for the controlled lab. For every major error family, include a visual example, its metric effect, a plausible mechanism, and one controlled intervention that could falsify the explanation.

Pathology bridge

Section titled “Pathology bridge”

Nuclei, glands, tissue regions, and rare lesions make boundary and object metrics clinically relevant; WSIs also introduce multiscale selection before dense prediction. Continue to cell, gland, and tissue models and pathology VLMs. Preserve physical magnification, patient-level splits, annotation provenance, and the distinction between class-agnostic proposal quality and diagnostic semantics.

Atlas-conditioned foundation segmentation and reliability fusion

A single annotated volume can generate spatial prompts and an independent registered-mask prediction. AtlasSegFM learns a voxel-wise fusion gate through support-space consistency while leaving the foundation backbone frozen. Read it as a one-shot supervision budget with substantial test-time adaptation.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Atlas is Your Perfect Context: One-Shot Customization for Generalizable Foundational Medical Image SegmentationZiyu Zhang, Yi Yu, Simeng Zhu, Ahmed Aly, Yunhe Gao, Ning Gu, and Yuan Xue2026 · ECCV · acceptedECCV 2026 Spotlight verified in the official program, session Computational, Medical & Scientific Imaging; scientific claims use conference-hosted manuscript #2028. Publisher proceedings/version of record was not verified on 2026-09-12.AtlasSegFM customizes a frozen segmentation foundation model using one annotated atlas. It registers atlas anatomy to the query, converts the warped mask into prompts, and combines atlas and foundation-model predictions with a voxel-wise reliability gate. One-shot refers to the support annotation budget, not absence of test-time optimization.ECCV Spotlight program · Conference manuscript

Reading question: What is labeled, what is optimized for each query, and can the gate be trained without using query ground truth?

Evidence boundary: Anatomical mismatch can corrupt both branches; interactive methods receiving ground-truth-derived clicks have a different information budget. CT/MRI anatomy is not WSI validation.

Readable prompt ensembles, local concepts and selective adaptation

TPB reweights mistakes across discrete prompt classifiers, CFM exposes spatially grounded sparse concepts, and Robotic Steering selects action-predictive attention heads. Compare what is selected or trained, how frozen features are reused, and whether the final intervention preserves unrelated capabilities.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
CFM: Language-aligned Concept Foundation Model for VisionKai Wittenmayer, Sukrut Rao, Amin Parchami-Araghi, Bernt Schiele, and Jonas Fischer2026 · ECCV · acceptedECCV 2026 Spotlight verified in the official Vision-Language Models & Visual Foundation Models session; manuscript #4902 checked on 2026-09-12. Publisher version of record was not independently verified.CFM converts dense language-aligned visual features into spatially grounded sparse concepts. A Matryoshka sparse autoencoder reconstructs patch features at nested dictionary capacities, separating coarse concepts from finer residual detail. Data-derived parent-child relationships then improve concept naming and support interpretable classification, captioning and open-vocabulary segmentation.ECCV Spotlight program · Conference manuscript · Official project or implementation

Reading question: Is the adaptation unit a prompt bank, a concept dictionary, an attention head, or another trainable branch?

Evidence boundary: Frozen backbones do not imply label-free or zero-cost adaptation. Probe predictability is not causal proof, and Robotic Steering also trains FFN LoRA and its action head.

Teacher-class mixtures and privileged residual routing

MoE-KD reformulates classification through latent teacher-class experts and EM. MARS uses full-minus-partial features to teach a deployable router. Specify whether knowledge resides in label mixtures, prototypes, routing decisions or a residual target.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Residual-Guided Expert Specialization for Incomplete Multimodal LearningSeunghun Baek, Jihwan Park, Jaeyoon Sim, Minjae Jeong, Hoseok Lee, Won Hwa Kim2026 · ECCV · acceptedECCV 2026 poster verified from the official linked program event. Review uses the conference-hosted manuscript checked 2026-09-12; a separate Springer version of record was not verified. MARS distills a privileged residual router into an available-feature router; it is not a cascade of residual-error student networks. Full modalities are assumed available during training. Food-101 full-modality accuracy is below SimMLM despite higher missing-modality and average scores.MARS uses the difference between complete- and partial-modality representations to teach expert routing. A privileged residual router learns which experts address each missingness pattern; a second router learns to imitate those decisions from available features alone. Thus the residual is a training signal, while deployment uses a practical feature router without the missing modalities.Official ECCV poster record · Official code

Reading question: Which privileged information is available only during training, and exactly what remains in the student?

Evidence boundary: Residual-guided routing is not a residual-error student cascade; conditional optimization arguments do not certify best accuracy.

Volume tokens, persistent geometry and registration

Compare learned volume representations, streaming geometric context, flow-based rigid registration and dense image matching. Each chooses a different correspondence unit and computation path; learned priors complement rather than abolish geometric constraints.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Volume Transformer: Revisiting Vanilla Transformers for 3D Scene UnderstandingKadir Yilmaz, Adrian Kruse, Tristan Höfer, Daan de Geus, and Bastian Leibe2026 · ECCV · acceptedAccepted to ECCV 2026; oral presentation verified in the official program. This record and review use the conference-hosted manuscript checked 2026-09-12; a separate Springer proceedings record was not verified.Volt makes a plain global-attention Transformer useful for 3D scene segmentation by turning occupied cubic volumes into tokens. Its central lesson is that tokenization and training data determine whether a generic architecture succeeds: a convolutional teacher supplies a useful inductive bias even when that teacher is less accurate than the trained student.Official ECCV oral presentation · Official ECCV paper record · arXiv record · Official code · Official project

Reading question: Which coordinates, features or correspondences are predicted, and how are rigidity, memory growth and distribution shift tested?

Evidence boundary: Compact per-frame context still accumulates; multi-step integration is not one-pass inference, and improved average matching can hide benchmark regressions.

Diffusion-boundary self-distillation and negative transfer

TRACE turns trajectory-derived attention boundaries into one-pass predictions through LoRA and an edge decoder. Read its natural-image gains alongside the reported MoNuSeg/TNBC degradation.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
TRACE: Your Diffusion Model Is Secretly an Instance Edge DetectorSanghyun Jo, Ziseok Lee, Wooyeol Lee, Jonghyun Choi, Jaesik Park, Kyungsu Kim2026 · ICLR · peer-reviewedOfficial ICLR 2026 oral and full proceedings manuscript checked. Diffusion-boundary self-distillation improves selected natural-image segmentation but degrades MoNuSeg/TNBC cell segmentation. One-pass speed is edge extraction only, with the large backbone retained. No WSI or cross-magnification validation; official code/checkpoints remain promised rather than released at the checked revision.TRACE extracts instance-boundary cues from self-attention inside a pretrained diffusion model, then self-distills a slower trajectory-based edge procedure into one backbone pass. Natural-image mask refinement improves, but the paper also reports worse cell segmentation on MoNuSeg and TNBC.OpenReview · Official project · Official repository — release pending

Reading question: Does the teacher prior delineate the target objects before it is distilled, and what remains in the deployed model?

Evidence boundary: A tiny decoder does not remove the large diffusion backbone. Edge-extraction speed is not end-to-end speed, and negative cell-level transfer is not WSI validation.

Conflicting supervision and adaptive distillation objectives

DTO-KD adapts the balance between task and teacher-transfer objectives using relative-loss optimization. Compare the exact log-gradient construction with the practical amortized update; control the transfer architecture before attributing gains to optimization.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
DTO-KD: Dynamic Trade-off Optimization for Effective Knowledge DistillationZeeshan Hayder, Ali Cheraghian, Lars Petersson, Mehrtash Harandi, Richard Hartley2026 · ICLR · peer-reviewedOfficial ICLR 2026 oral and proceedings manuscript checked. Adaptive task/distillation optimization, not cross-magnification or residual-error KD. Separate the exact two-gradient construction from the amortized implementation; the printed simplex handling and lower-bound argument require qualification, documented in the source-checked review.DTO-KD treats task supervision and teacher imitation as a multi-objective optimization problem. Its central idea is to adapt their contribution according to relative loss improvement instead of choosing one fixed task/distillation weight. The paper distinguishes conflicting gradient directions from unequal gradient magnitudes, but its exact analytical construction and practical amortized algorithm must not be treated as identical.Official proceedings record · OpenReview · Theoretical supplement

Reading question: Which objectives conflict, which update is actually implemented, and what costs and guarantees belong to that implementation?

Evidence boundary: The printed closed form omits simplex boundary handling and the claimed non-collapse lower bound has a counterexample. Natural-image results do not establish cross-magnification or WSI validity.

Channel distributions and feature-pyramid transfer

CWD normalizes spatial activations inside each channel; CSCWD adds teacher P2 to student P3 alignment for tiny-object detection. Study what the spatial probability distribution preserves before changing feature resolution.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Channel-Wise Knowledge Distillation for Dense PredictionChangyong Shu, Yifan Liu, Jianfei Gao, Zheng Yan, and Chunhua Shen2021 · ICCV · peer-reviewedICCV 2021 proceedings paper and author v4 manuscript checked; older missing prerequisite added during the October 2026 audit, not a new 2026 paper. Official repository records the semantic-segmentation code release on 20 August 2021.Channel-wise KD changes the axis on which dense-prediction activations become probabilities. Instead of normalizing classes separately at each pixel, it normalizes spatial positions separately within each channel. The student then matches teacher spatial distributions, encouraging it to preserve where each channel concentrates its evidence.Official implementation · Author manuscript
CSCWD: Cross-Scale Channel-wise Knowledge Distillation for Lightweight Tiny Object Detection on Edge DevicesAmir Zamani and Zeinab Ghasemi-Naraghi2026 · arXiv · preprintarXiv v1 checked 2 October 2026; no official code release or acceptance verified. Cross-scale means teacher P2 to student P3 feature-map alignment at a shared input resolution, not independent high/low magnification acquisitions. MGD is stronger on reported mAP; external stricter mAP does not improve.CSCWD lets a compact detector learn from a teacher feature level that the student does not possess. The teacher’s fine P2 map supervises the student’s coarser P3 map through training-only alignment and channel-wise distribution matching. Only the original YOLO11n inference graph remains at deployment.

Reading question: Are the two models seeing different acquisitions, or the same image through different feature maps?

Evidence boundary: CSCWD uses shared-resolution image inputs, not 20×/5× tissue pairs. Its recall gain coexists with MGD’s stronger mAP and a non-improving external strict mAP.

Tissue distillation: masks, features, hierarchy and relations

Compare staged pixel/feature KD from Virchow2, PAGET’s hierarchical multi-teacher pseudo-labels, and ConStruct’s transfer of SegFormer structural affinities into adapted CONCH features for prototype-guided segmentation. Their annotation contracts and deployed components differ.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation ModelsJiaqi Lv, Yijie Zhu, Saki Okada, Maxwell Stanley Renna, Robert Goldin, Stefan Antonowicz, and Shan E Ahmed Raza2026 · arXiv · preprintarXiv v1 (2026-09-03), submitted to Medical Image Analysis according to the authors, not accepted. Verified methods and selected results on 2026-10-02. TIAToolbox is the stated release destination; availability of these exact weights was not independently confirmed. Patch-throughput measurements are not end-to-end WSI latency.Transfer Virchow2 segmentation knowledge into SegFormerB5 or EUNet-M using a staged combination of pixel predictions and global features. The deployed model is a compact segmentation student, not the foundation-model teacher. This changes model cost at fixed patch input, not optical magnification.Versioned manuscript · TIAToolbox — stated release destination

Reading question: Which supervision remains observable at student resolution, and which teacher or post-processing components still run at inference?

Evidence boundary: PAGET-H retains HoverNet. ConStruct’s residual adapters are not residual-error targets. Pixel segmentation, patch throughput and patient-level WSI outcomes are distinct evidence.

Global semantic anchors and local patch targets

Two Global Crops Suffice separates same-image global-view alignment from local crops and iBOT refinement in controlled DINO-family experiments. CODI instead changes the patch target: weakly mix the same teacher view’s class distribution into its local patch distribution. The diagnostic study and target-side method answer related questions, but neither independently validates the other.

Primary paper · year · evidenceCore idea and why to read itOfficial resources
CODI: Improving Self-Supervised Vision Transformers with Cross DistillationSiran Dai, Qianqian Xu, Peisong Wen, Yang Liu, Xiaochun Cao, and Qingming Huang2026 · arXiv · preprintarXiv:2610.30639v1, submitted 22 October 2026, inspected 26 October 2026. The manuscript has a NeurIPS 2026 footer, but no official acceptance record was independently located in this audit; retained as a preprint. The author repository contains only a Coming soon README, not released implementation code.CODI changes a masked patch’s teaching target, not the Vision Transformer architecture. It mixes 90% of the momentum teacher’s local patch distribution with 10% of that same view’s global class-token distribution. This gives low-attention patches a direct semantic training signal while retaining mostly local supervision. It is self-distillation during unlabeled image pretraining, not compression from a separately pretrained large teacher and not cross-magnification transfer.Inspected manuscript (v1) · Author repository — implementation pending at inspection

Reading question: Which signal creates useful spatial semantics, which refines it, and do your global and dense evaluations agree?

Evidence boundary: Both inspected sources are preprints. Crop ablations match iterations, not all computation; the depth-scale narrative has an appendix exception. CODI’s gradient statement is local/logit-level, its alpha is task-dependent, and its code was not released at inspection. Neither reports WSI or cross-magnification validation.

Complete maintained paper route

Section titled “Complete maintained paper route”

51 canonical readings · 2014–2026· route reviewed through October 2, 2026.

This is the complete structured list for this chapter, ordered chronologically from earlier foundations toward current work. The explanatory ladder above highlights the shortest conceptual route; this list also includes important branches, benchmarks, and current endpoints. It is generated from the same deduplicated catalog used by the AI background registry and My Library, so a future catalog update appears here automatically.

Show all 51 readings in maintained order

  2. #22015ICCVPeer reviewedL1–L2

    Fast R-CNN

    Ross Girshick

    Metadata checked

  12. #122017ICCVPeer reviewedL2

    Mask R-CNN

    Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick

    Metadata checked

  23. #232021IEEE Transactions on Image ProcessingPeer reviewedL3

    Resolution-Aware Knowledge Distillation for Efficient Inference

    Zhanxiang Feng, Jianhuang Lai, and Xiaohua Xie

    Peer-reviewed journal article, volume 30, pages 6985–6996. The paper calls its framework RKD, meaning resolution-aware—not relational—knowledge distillation. Teacher and student use the same architecture but paired high- and low-resolution inputs; only the low-resolution student is needed at inference.

    Official artifacts: PubMed record

    Metadata checked

  32. #322023ICCVPeer reviewedL2–L3

    Segment Anything

    Alexander Kirillov et al.

    Metadata checked

  42. #422026ECCVAccepted · proceedings pendingL3

    Volume Transformer: Revisiting Vanilla Transformers for 3D Scene Understanding

    Kadir Yilmaz, Adrian Kruse, Tristan Höfer, Daan de Geus, and Bastian Leibe

    Accepted to ECCV 2026; oral presentation verified in the official program. This record and review use the conference-hosted manuscript checked 2026-09-12; a separate Springer proceedings record was not verified.

    Official artifacts: Official ECCV oral presentation · Official ECCV paper record · arXiv record · Official code · Official project

    Metadata checked

  43. #432026ECCVAccepted · proceedings pendingL3

    Atlas is Your Perfect Context: One-Shot Customization for Generalizable Foundational Medical Image Segmentation

    Ziyu Zhang, Yi Yu, Simeng Zhu, Ahmed Aly, Yunhe Gao, Ning Gu, and Yuan Xue

    ECCV 2026 Spotlight verified in the official program, session Computational, Medical & Scientific Imaging; scientific claims use conference-hosted manuscript #2028. Publisher proceedings/version of record was not verified on 2026-09-12.

    Official artifacts: ECCV Spotlight program · Conference manuscript

    Metadata checked

  44. #442026ECCVAccepted · proceedings pendingL3

    CFM: Language-aligned Concept Foundation Model for Vision

    Kai Wittenmayer, Sukrut Rao, Amin Parchami-Araghi, Bernt Schiele, and Jonas Fischer

    ECCV 2026 Spotlight verified in the official Vision-Language Models & Visual Foundation Models session; manuscript #4902 checked on 2026-09-12. Publisher version of record was not independently verified.

    Official artifacts: ECCV Spotlight program · Conference manuscript · Official project or implementation

    Metadata checked

  45. #452026ECCVAccepted · proceedings pendingL3

    Residual-Guided Expert Specialization for Incomplete Multimodal Learning

    Seunghun Baek, Jihwan Park, Jaeyoon Sim, Minjae Jeong, Hoseok Lee, Won Hwa Kim

    ECCV 2026 poster verified from the official linked program event. Review uses the conference-hosted manuscript checked 2026-09-12; a separate Springer version of record was not verified. MARS distills a privileged residual router into an available-feature router; it is not a cascade of residual-error student networks. Full modalities are assumed available during training. Food-101 full-modality accuracy is below SimMLM despite higher missing-modality and average scores.

    Official artifacts: Official ECCV poster record · Official code

    Metadata checked

  46. #462026ICLRPeer reviewedL3

    DTO-KD: Dynamic Trade-off Optimization for Effective Knowledge Distillation

    Zeeshan Hayder, Ali Cheraghian, Lars Petersson, Mehrtash Harandi, Richard Hartley

    Official ICLR 2026 oral and proceedings manuscript checked. Adaptive task/distillation optimization, not cross-magnification or residual-error KD. Separate the exact two-gradient construction from the amortized implementation; the printed simplex handling and lower-bound argument require qualification, documented in the source-checked review.

    Official artifacts: Official proceedings record · OpenReview · Theoretical supplement

    Metadata checked

  47. #472026ICLRPeer reviewedL3

    TRACE: Your Diffusion Model Is Secretly an Instance Edge Detector

    Sanghyun Jo, Ziseok Lee, Wooyeol Lee, Jonghyun Choi, Jaesik Park, Kyungsu Kim

    Official ICLR 2026 oral and full proceedings manuscript checked. Diffusion-boundary self-distillation improves selected natural-image segmentation but degrades MoNuSeg/TNBC cell segmentation. One-pass speed is edge extraction only, with the large backbone retained. No WSI or cross-magnification validation; official code/checkpoints remain promised rather than released at the checked revision.

    Official artifacts: OpenReview · Official project · Official repository — release pending

    Metadata checked

  48. #482026arXivPreprintL3

    Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation Models

    Jiaqi Lv, Yijie Zhu, Saki Okada, Maxwell Stanley Renna, Robert Goldin, Stefan Antonowicz, and Shan E Ahmed Raza

    arXiv v1 (2026-09-03), submitted to Medical Image Analysis according to the authors, not accepted. Verified methods and selected results on 2026-10-02. TIAToolbox is the stated release destination; availability of these exact weights was not independently confirmed. Patch-throughput measurements are not end-to-end WSI latency.

    Official artifacts: Versioned manuscript · TIAToolbox — stated release destination

    Metadata checked

  50. #502026arXivPreprintL3

    CODI: Improving Self-Supervised Vision Transformers with Cross Distillation

    Siran Dai, Qianqian Xu, Peisong Wen, Yang Liu, Xiaochun Cao, and Qingming Huang

    arXiv:2610.30639v1, submitted 22 October 2026, inspected 26 October 2026. The manuscript has a NeurIPS 2026 footer, but no official acceptance record was independently located in this audit; retained as a preprint. The author repository contains only a Coming soon README, not released implementation code.

    Official artifacts: Inspected manuscript (v1) · Author repository — implementation pending at inspection

    Metadata checked