Skip to content

CVPR 2026 — Program Notes (pre-meeting prep)

Last updated: May 11, 2026 — T-3.5 weeks. Final paper decisions landed Feb 20, 2026; accepted-paper list public on OpenReview; CVF Open Access PDFs post around the meeting. Workshop accepted-paper lists landing May–early June. Camera-ready in flight.

This page exists to lock the scope filter, workshop slate, anticipated paper themes, and cross-vault pairings before any tool extraction begins. It is intentionally explicit about what is out of scope — CVPR's bulk is general computer vision, and conflating that bulk with the medical-imaging / pathology-FM subset would destroy the vault's signal-to-noise.

Verified meeting facts

  • Dates: Wed June 3 – Sun June 7, 2026.
  • Jun 3 (Wed): Workshops + Tutorials Day 1 — the medical cluster lives here (MCV full day, FMV half day, MMFM-BIOMED half day, PHAROS half day).
  • Jun 4 (Thu): Workshops + Tutorials Day 2 — second medical sub-cluster (CV4Clinical half day, Med-Reasoner half day).
  • Jun 5 (Fri) – Jun 7 (Sun): Main conference, orals + highlights + poster sessions.
  • Venue: Colorado Convention Center, Denver, Colorado, USA.
  • Scale: ~10,000 attendees (sibling to NeurIPS / ICML in scale).
  • Proceedings: CVF Open Access (full open-access PDFs, no paywall) — typically posts within ~1 week of meeting. OpenReview venue: CVPR.thecvf.com/2026/Conference.

Scope filter (load-bearing — repeat from tools/index.md)

CVPR is scope-limited by design in this vault. The filter is:

  1. Medical / biomedical workshop (one of six identified).
  2. Main-track keyword match on medical / pathology / radiology / oncology / clinical / biomedical / segmentation-for-medicine vocabulary.
  3. Architectural building block with a documented or near-certain medical adoption path — DINOv3-class SSL pretraining, SAM-3-class segmentation FM, next-gen CLIP / SigLIP multimodal alignment, slide-level aggregation, diffusion / flow-matching for medical synthesis.

Everything else stays out. The light-build pattern mirrors ICML 2026 and NeurIPS 2026.

Medical / biomedical workshop slate (verified)

Six workshops identified from the CVPR 2026 workshops page. All run on Jun 3–4 alongside tutorials.

Workshop Acronym Date Format Contact Notes
Medical Computer Vision MCV Wed Jun 3 Full day Zongwei Zhou 12th edition; the canonical CVPR medical-CV workshop; broad scope from segmentation through FMs
Foundation Models for Medical Vision FMV Wed Jun 3 Half day Jun Ma 3rd edition; medical-imaging FMs, SAM-Med-class, RadFM-class; co-organized with the BraTS / nnU-Net network
Multimodal Foundation Models for Biomedicine MMFM-BIOMED Wed Jun 3 Half day Yuhui Zhang 2nd edition; pathology + radiology VLMs; CLIP / LLaVA / SigLIP for biomedicine
PHAROS AI Factory for Medical Imaging & Healthcare PHAROS Wed Jun 3 Half day Stefanos Kollias EU AI Factory project; healthcare deployment, regulatory, federated learning
Bridging AI and Medical Reality (CV4Clinical) CV4Clinical Thu Jun 4 Half day Yicheng Wu Real-world clinical translation, reliability, reproducibility, deployment failure modes
Medical Reasoning with Vision Language Foundation Models Med-Reasoner Thu Jun 4 Half day Anas Zafar New entrant for 2026; reasoning capabilities for medical VLMs; clinician × CV researcher cross

Workshop accepted-paper lists are landing through May 2026; stub pages and per-paper tracking will populate as those lists go public.

Anticipated paper themes (5 high-signal axes)

These are the patterns we expect to see repeatedly in both main-track and workshop accepts. We will not commit to a full theme synthesis page (that would be out of scope for a light build), but the per-tool entries will tag against these themes:

  1. Pathology foundation model successors and slide-level aggregators. Next-generation tile encoders (DINOv3-class, scaling beyond UNI / GigaPath / Virchow), slide-level aggregation transformers (ABMIL / TransMIL successors), and multi-cohort pretraining at TCGA / CPTAC / Mass General Brigham + Providence + UCSF-scale. Direct upstream of CHIEF / Prov-GigaPath / UNI updates.

  2. Pathology and radiology vision-language models. CONCH / PathChat / MUSK successors on the pathology side; MAIRA / RadFM / Merlin / CXR-LLaVA successors on the radiology side. Pathology-report alignment, radiology-report generation, prompt-tunable report drafting. Expect MMFM-BIOMED + Med-Reasoner to be the workshop home for the bulk of these.

  3. Generative medical imaging and counterfactuals. Latent diffusion / DiT / flow-matching priors for MRI / CT reconstruction; ControlNet-style anatomy-conditioned synthesis; counterfactual generation for fairness audits and explainability; synthetic-data augmentation for rare-cohort segmentation. Bridges to MICCAI 2026 DGM4MICCAI workshop.

  4. Segmentation foundation models — SAM line and successors. SAM-3 / EfficientSAM-2 / MedSAM-class adapters; promptable segmentation for tumor / lesion / organ tasks; few-shot and zero-shot transfer to BraTS / HECKTOR / KiTS / AutoPET / PANORAMA. Direct upstream of MICCAI challenge entries.

  5. Multimodal medical reasoning and clinical-grade VLMs. Med-Reasoner is the named workshop; the main track will carry the architecture papers. Chain-of-thought reasoning over chest X-rays / pathology slides, agentic clinical workflows, retrieval-augmented diagnosis, multimodal alignment between text reports + imaging + structured EHR. Cross-link to AACR 2026 agentic-AI axis and ICML 2026 agentic-AI subset.

Out-of-scope themes (explicitly skipped)

To keep the filter honest, the following CVPR 2026 themes get no coverage in this vault even when they are technically impressive:

  • General scene understanding, depth estimation, optical flow, monocular 3D — no medical thread.
  • Autonomous driving perception (KITTI / Waymo / nuScenes-class) — no medical thread.
  • Robotics manipulation, embodied AI, simulation-to-real — no medical thread (surgical robotics is MICCAI's CAI track, not CVPR's general robotics track).
  • General video generation (Sora / Veo / Movie Gen class) — no medical thread unless medical-imaging-specific.
  • Face recognition, person re-identification, biometrics — no medical thread.
  • AR / VR / NeRF / Gaussian splatting (general) — no medical thread.
  • Adversarial robustness and watermarking for general images — out of scope.
  • General efficiency / quantization / pruning — out of scope unless the paper benchmarks on a medical FM.

Cross-vault pairings (the load-bearing chains)

The point of carrying CVPR in this vault is the cross-conference arcs. Three canonical chains to track:

  1. CVPR pretraining recipe → MICCAI methods → AACR translational poster. A new SSL recipe at CVPR June 2026 → pathology-FM v2 paper at MICCAI 2026 Sep 27–Oct 1 → translational poster at AACR 2027 April. The CVPR entry should explicitly link forward to the MICCAI cousin and the eventual AACR poster.

  2. CVPR medical-VLM → RSNA clinical-deployment readout. A new radiology VLM at CVPR / MMFM-BIOMED → vendor demonstration or prospective-study readout at RSNA 2026 Nov 29–Dec 3. The half-year arc is reliable for diagnostic-imaging VLMs.

  3. CVPR SAM successor → MICCAI challenge winner. A SAM-3-class segmentation FM at CVPR → adoption in BraTS / HECKTOR / KiTS / AutoPET / PANORAMA winning entry at MICCAI 2026 within the same calendar year. Track which CVPR segmentation paper anchors each MICCAI challenge result.

What is NOT being built

To be explicit (light-build hygiene):

  • No digest/ directory. No per-day digests. The general-vision bulk dominates per-day proceedings and would dilute signal.
  • No themes.md. The five themes above are scoped here in program-notes.md and tagged per tool entry; a separate synthesis page is not warranted.
  • No keynotes/. CVPR plenaries are general-vision; we cover them only if a known medical-imaging / pathology-FM author is on stage.
  • No mkdocs.yml. This vault piggybacks on the parent site config (per repo convention for sibling conference vaults).
  • No timeline.md. Conference timeline lives at ../../timeline.md; no per-vault duplication.

Next actions

  • May 11–31, 2026 (now → T-1 week): scan accepted-paper list with keyword filter; pre-stub the ~25–50 high-signal entries. Pull workshop accept lists as they post. Cross-check against the MICCAI 2026 early-accept list (already posted May 7) for cousin-paper pairing.
  • Jun 3–7, 2026 (during the meeting): opportunistic capture of slide deposits, GitHub repos, demo links, social reactions. The Wed Jun 3 medical-workshop cluster (MCV / FMV / MMFM-BIOMED / PHAROS) is the densest day; Thu Jun 4 carries the clinical-translation pair (CV4Clinical + Med-Reasoner).
  • Jun 8 – Jun 30, 2026 (post-meeting): CVF Open Access proceedings post ~1 week after meeting. Bulk-populate tools/ with the filtered subset. Cross-link into AACR pathology-FM dossiers and stage MICCAI 2026 (Sep 27–Oct 1) cousin pairings.

Sources