Pith. sign in

REVIEW 3 major objections 6 minor 38 references

NPCs learn to see their world and talk about it

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

2026-07-05 09:11 UTC pith:CAPME2KJ

load-bearing objection NPC dialogue paper has a title/abstract mismatch with a medical imaging paper — the actual content is a modest systems paper with no real evaluation the 3 major comments →

arxiv 2604.19191 v2 pith:CAPME2KJ submitted 2026-04-21 cs.CV cs.AI

Towards Modality-Agnostic Medical Image Anomaly Detection: A Training-Free Manifold Refinement Approach

Pritam Kar , Gouri Lakshmi S , Saptarshi Bej This is my paper
classification cs.CV cs.AI
keywords acrossanomalydatasetsdensityframeworkmethodsrefinementbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a system that gives non-playable characters (NPCs) in games real-time awareness of their surroundings by capturing panoramic images at the NPC's location, running semantic segmentation to identify objects and their spatial positions, and feeding this structured environmental data—along with scene graph information encoded as directional vectors—into a large language model (LLM). The LLM then generates dialogue that references nearby objects, landmarks, and spatial relationships, rather than relying on pre-scripted dialogue trees. The system is implemented as a plugin for Unreal Engine 5 and uses the Recognize Anything Model (RAM) for open-set object recognition. An expert interview and a user study both indicate that players preferred the context-aware NPC over a baseline LLM-based NPC without environmental grounding. The core argument is that by converting visual perception of the game world into a structured text representation, one can bridge the gap between static dialogue systems and NPCs that dynamically reflect their environment in conversation.

Core claim

The central mechanism is a pipeline that converts panoramic images of a game scene into a structured JSON representation containing segmented object labels, their spatial positions as directional vectors, and scene graph data within the NPC's bounding sphere. This structured text is injected into the LLM's prompt, enabling the NPC to ground its generated dialogue in real-time environmental perception without any retraining of the language model itself. The contribution is showing that this post-hoc grounding—layering perception output onto an existing LLM—is sufficient to produce dialogue that users perceive as more contextually appropriate and engaging than a non-context-aware baseline.

What carries the argument

The pipeline consists of: (1) panoramic image capture at the NPC's location in Unreal Engine 5, (2) semantic segmentation using the Recognize Anything Model (RAM) to identify objects, (3) extraction of object positions as directional vectors within a bounding sphere around the NPC, (4) composition of all extracted data into a structured JSON prompt, and (5) submission of this prompt to a GPT-4 API to initialize the NPC's environmental context before player interaction begins.

Load-bearing premise

The paper assumes that a structured JSON representation of the environment—object labels plus coarse directional vectors—is rich enough for the LLM to generate spatially grounded, contextually appropriate dialogue. If the environmental representation loses too much spatial detail (e.g., object-to-object relationships, depth, or fine-grained positioning), the LLM's responses may be generic or hallucinate environmental details.

What would settle it

If users in a blinded comparison could not reliably distinguish between an NPC given the full environmental context and one given only a generic supporting prompt with no environmental data, the value of the perception pipeline would be undermined.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the approach generalizes, any game engine with access to rendered views could give NPCs environmental awareness without custom AI training, lowering the barrier for indie developers.
  • The same pipeline could be applied to virtual training simulations, museum guides, or accessibility tools where a conversational agent needs to reference its physical surroundings.
  • Adding depth information (via depth rendering) and finer directional quantization (8 or 16 directions) could substantially improve spatial reasoning quality, as the authors themselves note.
  • The approach could be extended beyond static panoramic capture to continuous real-time perception, allowing NPCs to comment on dynamic events as they unfold.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The manuscript submitted under arXiv:2604.19191 (cs.CV) with the title 'Towards Modality-Agnostic Medical Image Anomaly Detection: A Training-Free Manifold Refinement Approach' presents a critical metadata mismatch: the abstract describes a training-free framework for medical image anomaly detection using manifold refinement, but the full text is an entirely different paper titled 'Empowering NPC Dialogue with Environmental Context Using LLMs and Panoramic Images,' a technical report on game AI. The full-text paper proposes enhancing non-playable characters (NPCs) in games by combining large language models (LLMs) with panoramic image semantic segmentation and scene graph data. The system captures panoramic images of an NPC's environment, applies semantic segmentation (using RAM), extracts object locations and directional vectors within a bounding sphere, and composes a structured JSON prompt for an LLM (GPT-4). Evaluation consists of an expert interview (n=1, single scene, four input configurations) and a user study comparing context-aware NPCs against a baseline. The central claim is that integrating environmental context via panoramic segmentation and scene graphs enables NPCs to produce more believable and engaging dialogue.

Significance. The full-text paper addresses an interesting and relatively underexplored intersection of computer vision, game AI, and natural language processing. The idea of grounding LLM-based NPC dialogue in real-time environmental perception using panoramic segmentation is a reasonable contribution to the emerging area of LLM-driven game AI. However, the significance of the contribution is severely undermined by the abstract–full-text mismatch, which makes it impossible to assess the paper as submitted. The evaluation methodology (expert interview with n=1, user study with no reported sample size or quantitative metrics) does not meet the standards of the field for supporting claims of user preference or engagement. The system does not ship reproducible code, machine-checked proofs, or parameter-free derivations; the two free parameters (bounding sphere radius, directional vector quantization) are noted but not systematically evaluated.

major comments (3)
  1. Abstract–full-text mismatch: The abstract supplied for review describes a medical image anomaly detection framework ('manifold refinement' for OCC on the MedIAnomaly benchmark), while the full text is a technical report on NPC dialogue systems using LLMs and panoramic images. This mismatch is load-bearing because it makes the paper's actual contribution and scope impossible to verify as submitted. The authors must clarify which paper is the actual submission and ensure the abstract, title, and metadata are consistent before any substantive assessment can proceed.
  2. Section 4, Evaluation: The central claim that context-aware NPCs are 'more believable and engaging' rests on a user study with no reported sample size, no quantitative preference metrics, no statistical tests, and no blinding protocol. The text states participants 'preferred' the context-aware NPCs but provides no numerical data. Without knowing how many participants were involved, whether the comparison was blinded, what the effect size was, or whether the result is statistically significant, the preference claim is anecdotal and cannot support the paper's claims.
  3. Section 4.1, Expert Interview: The expert interview uses n=1 expert on a single scene with four input configurations. While this is described as a preliminary evaluation, the paper draws conclusions about system effectiveness from this. The expert feedback (e.g., preference for shorter answers, desire for finer direction decimation) is qualitative and not systematically analyzed. This evaluation is insufficient as primary evidence for the claims made.
minor comments (6)
  1. Throughout Sections 3.1, 3.5, 3.7, and 4: Large portions of the text appear as garbled or placeholder characters (e.g., 'supporting prompt,' 'structured JSON representation,' and NPC dialogue examples are rendered as sequences of replacement characters). This renders substantial portions of the manuscript unreadable and must be fixed for reproducibility and review.
  2. Section 3: The method description is incomplete. Subsections referenced in the text (e.g., Section 3.3, 3.4, 3.5) are either missing or truncated in the provided text. The pipeline stages (Bounding Sphere Object Selection, Radial Object Selection, Input Composition) are mentioned but not fully described, making it difficult to assess technical sufficiency.
  3. Section 3.1: The system relies on GPT-4 via API, RAM for segmentation, and Unreal Engine 5. No details are provided on API parameters (temperature, max tokens), RAM model version, or UE5 plugin configuration. These details affect reproducibility.
  4. The free parameters noted in the axiom ledger (bounding sphere radius, directional vector quantization) are not explicitly identified or justified in the main text. The paper mentions 'maybe 8 or 16 directions' as future work but does not report the value actually used.
  5. References: Several citations appear with garbled or missing characters in titles and venue names (e.g., Isbister [2022], Laird [2001], Feng [2014]). Reference formatting should be corrected.
  6. The paper is labeled 'TECHNICAL REPORT' and the arXiv identifier (2604.19192v2) differs from the one in the review assignment (2604.19191). The submission category (cs.GR vs. cs.CV) should be clarified and metadata adjusted accordingly.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. The referee raises three major points, all of which are valid. We address each below.

read point-by-point responses

  1. Referee: Abstract–full-text mismatch: The abstract describes a medical image anomaly detection framework while the full text is about NPC dialogue systems using LLMs and panoramic images.

    Authors: The referee is entirely correct. There is a metadata mismatch between the abstract/title supplied for review and the actual full-text manuscript. The full text is our intended submission: 'Empowering NPC Dialogue with Environmental Context Using LLMs and Panoramic Images.' The abstract describing medical image anomaly detection was erroneously attached during the submission process. We will correct the abstract, title, and all metadata to be consistent with the full-text paper before any resubmission. We acknowledge that this error made substantive assessment of the paper as submitted impossible, and we apologize for the confusion. revision: yes

  2. Referee: Section 4, Evaluation: The user study has no reported sample size, no quantitative preference metrics, no statistical tests, and no blinding protocol.

    Authors: The referee's assessment is correct. The current manuscript does not report the user study sample size, quantitative preference metrics, statistical tests, or a blinding protocol. This is a significant weakness in the evaluation. In the revised manuscript, we will: (1) report the exact number of participants, (2) include quantitative preference data (e.g., counts or ratings per condition), (3) apply appropriate statistical tests (e.g., binomial test or Wilcoxon signed-rank depending on the data structure), and (4) describe the blinding procedure used to ensure participants could not identify which condition was which. We will also report effect sizes. If the data collected do not support these analyses, we will conduct an additional, properly designed user study and report the results transparently. revision: yes

  3. Referee: Section 4.1, Expert Interview: n=1 expert on a single scene with four input configurations; qualitative feedback not systematically analyzed; insufficient as primary evidence.

    Authors: We agree that the expert interview with n=1 on a single scene cannot serve as primary evidence for the system's effectiveness. In the revised manuscript, we will reframe the expert interview explicitly as a formative pilot study whose purpose was to inform system refinements prior to the user study, not as confirmatory evidence. We will also temper the conclusions drawn from this component accordingly. Additionally, we will expand the evaluation to include multiple scenes and, if feasible, multiple expert reviewers to provide a more robust qualitative assessment. We acknowledge that the current framing overstates the evidentiary weight of this component. revision: yes

Circularity Check

0 steps flagged

No circularity found: the paper is an applied systems paper with no derivation chain to audit

full rationale

This paper describes an applied system that integrates panoramic image segmentation, scene graph data, and LLMs to provide environmental context for NPCs in games. There is no mathematical derivation chain, no fitted-parameter-to-prediction loop, no self-citation-based uniqueness theorem, and no ansatz smuggled through citation. The system feeds segmentation output and scene graph data as a structured JSON prompt to GPT-4, and the evaluation consists of an expert interview and a user study comparing context-aware vs. baseline NPCs. The central claim (that context-aware NPCs are preferred) rests on external evaluation, not on a circular derivation. The paper relies entirely on external tools (GPT-4, RAM/SAM, Unreal Engine 5) and does not define any quantity in terms of the result it claims to predict. While the evaluation has notable rigor gaps (no sample size, no quantitative metrics, no statistical tests), these are correctness/evaluation concerns, not circularity. The derivation chain is effectively: capture panoramic image → run semantic segmentation → extract object positions and scene graph → compose JSON prompt → send to LLM → return response. No step in this pipeline is defined in terms of its own output. The paper is self-contained against external benchmarks in the sense that its claims are externally falsifiable (users could in principle prefer the baseline), and no self-citation chain is load-bearing for the central result.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The paper does not introduce new mathematical entities or physical particles. It relies on existing software components and models. The free parameters are implementation choices for the game environment.

free parameters (2)
  • Bounding sphere radius
    The radius of the bounding sphere used to select objects for the scene graph is a parameter chosen for the implementation.
  • Directional vector quantization = 4
    The paper mentions using 4 directions, with a suggestion for future work to use 8 or 16, indicating this was a chosen parameter.
axioms (2)
  • domain assumption GPT-4 can generate contextually appropriate dialogue given structured environmental data.
    The system's core functionality relies on the LLM's ability to interpret and use the provided JSON data effectively.
  • domain assumption Semantic segmentation on panoramic images provides sufficient environmental detail for NPC interaction.
    The paper assumes that the output of the segmentation model is rich enough to drive meaningful context-aware responses.

pith-pipeline@v1.1.0-glm · 24458 in / 1504 out tokens · 166415 ms · 2026-07-05T09:11:37.463361+00:00 · methodology

0 comments
read the original abstract

Deploying AI-based anomaly detection across diverse clinical imaging settings remains challenging because most existing methods rely on modality-specific architectures, anatomical priors, or extensive retraining, limiting their use as general-purpose screening tools. One-class classification (OCC) offers a label-efficient alternative by training exclusively on normal data, but conventional two-stage pipelines fit a density estimator directly on raw pretrained embeddings, leaving substantial discriminative structure in the latent space unexploited. We introduce a training-free, modality-agnostic framework that inserts an explicit manifold-refinement stage between feature extraction and anomaly scoring. Empirical density weights, estimated via a UMAP-derived neighborhood graph, guide an iterative shift of embeddings toward locally dense regions, compacting normal samples, leaving anomalies relatively isolated prior to Gaussian density estimation and Mahalanobis-based scoring. This refinement introduces no additional trainable parameters and no architectural modification, allowing it to be layered onto any pretrained encoder. Evaluated on the MedIAnomaly benchmark across seven datasets spanning five imaging modalities (X-ray, MRI, fundus, dermatoscopy, histopathology), the framework achieves the best AUC on four datasets and the best Average Precision on five datasets among methods evaluated in the benchmark, outperforming specialized reconstruction and diffusion-based methods with a single fixed hyperparameter configuration across all modalities. These results demonstrate that meaningful gains can be achieved through post-hoc geometric refinement of existing representations rather than bespoke encoders, offering a practical and scalable AI screening framework for real-world, multi-modality clinical workflows where retraining and abnormal-case annotation are costly or infeasible.

Figures

Figures reproduced from arXiv: 2604.19191 by Gouri Lakshmi S, Pritam Kar, Saptarshi Bej.

Figure 1
Figure 1. Figure 1: Overview of the proposed MSDE anomaly detection pipeline. (a) Medical images are [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of UMAP embeddings for the RSNA (AnatPaste backbone) and C16 [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 5 internal anchors

  1. [1]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2623--2631, 2019

    work page 2019

  2. [2]

    The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification

    Ujjwal Baid, Suyash Ghodasara, Suyash Mohan, Michel Bilello, Evan Calabrese, Errol Colak, et al. The rsna-asnr-miccai brats 2021 benchmark on brain tumor segmentation and radiogenomic classification. arXiv preprint arXiv:2107.02314, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  3. [3]

    Autoencoders for unsupervised anomaly segmentation in brain mr images: A comparative study

    Christoph Baur, Stefan Denner, Benedikt Wiestler, Nassir Navab, and Shadi Albarqouni. Autoencoders for unsupervised anomaly segmentation in brain mr images: A comparative study. Medical Image Analysis, 69: 0 101952, 2021

    work page 2021

  4. [4]

    Laplacian eigenmaps for dimensionality reduction and data representation

    Mikhail Belkin and Partha Niyogi. Laplacian eigenmaps for dimensionality reduction and data representation. Neural Computation, 15 0 (6): 0 1373--1396, 2003

    work page 2003

  5. [5]

    Multidimensional binary search trees used for associative searching

    Jon Louis Bentley. Multidimensional binary search trees used for associative searching. Communications of the ACM, 1975

    work page 1975

  6. [6]

    Mask, Stitch, and Re-Sample: Enhancing Robustness and Generalizability in Anomaly Detection through Automatic Diffusion Models

    CI Bercea et al. Mask, stitch, and re-sample: Enhancing robustness and generalizability in anomaly detection through automatic diffusion models. arXiv preprint arXiv:2305.19643, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  7. [7]

    Uninformed students: Student-teacher anomaly detection with discriminative latent embeddings

    Paul Bergmann, Michael Fauser, David Sattlegger, and Carsten Steger. Uninformed students: Student-teacher anomaly detection with discriminative latent embeddings. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4183--4192, 2020

    work page 2020

  8. [8]

    Medianomaly: A comparative study of anomaly detection in medical images

    Yu Cai, Weiwen Zhang, Hao Chen, and Kwang-Ting Cheng. Medianomaly: A comparative study of anomaly detection in medical images. Medical Image Analysis, 102: 0 103500, 2025

    work page 2025

  9. [9]

    Medianomaly: Benchmark for medical image anomaly detection

    Yu Cai et al. Medianomaly: Benchmark for medical image anomaly detection. https://github.com/caiyu6666/MedIAnomaly, 2023

    work page 2023

  10. [10]

    Skin Lesion Analysis Toward Melanoma Detection 2018: A Challenge Hosted by the International Skin Imaging Collaboration (ISIC)

    Noel Codella, Veronica Rotemberg, Philipp Tschandl, M Emre Celebi, Stephen Dusza, David Gutman, et al. Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (isic). arXiv preprint arXiv:1902.03368, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  11. [11]

    The mahalanobis distance

    Roy De Maesschalck, Delphine Jouan-Rimbaud, and D \'e sir \'e L Massart. The mahalanobis distance. Chemometrics and intelligent laboratory systems, 50 0 (1): 0 1--18, 2000

    work page 2000

  12. [12]

    Padim: A patch distribution modeling framework for anomaly detection and localization

    Thomas Defard, Aleksandr Setkov, Angelique Loesch, and Romaric Audigier. Padim: A patch distribution modeling framework for anomaly detection and localization. In International Conference on Pattern Recognition, pages 475--489. Springer, 2021

    work page 2021

  13. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 770--778, 2016

    work page 2016

  14. [14]

    Jolliffe

    Ian T. Jolliffe. Principal Component Analysis. Springer, 2016

    work page 2016

  15. [15]

    Anomaly detection via mean shift density enhancement

    Pritam Kar, Rahul Bordoloi, Olaf Wolkenhauer, and Saptarshi Bej. Anomaly detection via mean shift density enhancement. arXiv preprint arXiv:2602.03293, 2026

    work page arXiv 2026

  16. [16]

    Antanas Kascenas, Nicolas Pugeault, and Alison Q. O'Neil. Denoising autoencoders for unsupervised anomaly detection in brain mri. In Proceedings of the 5th International Conference on Medical Imaging with Deep Learning, volume 172 of Proceedings of Machine Learning Research, pages 653--664. PMLR, 2022. URL https://proceedings.mlr.press/v172/kascenas22a.html

    work page 2022

  17. [17]

    Auto-Encoding Variational Bayes

    Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  18. [18]

    Cutpaste: Self-supervised learning for anomaly detection and localization

    Chun-Liang Li, Kihyuk Sohn, Jinsung Yoon, and Tomas Pfister. Cutpaste: Self-supervised learning for anomaly detection and localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9664--9674, 2021

    work page 2021

  19. [19]

    Attention based glaucoma detection: A large-scale database and cnn model

    Liu Li, Mai Xu, Xiaofei Wang, Lai Jiang, and Hanruo Liu. Attention based glaucoma detection: A large-scale database and cnn model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10571--10580, 2019

    work page 2019

  20. [20]

    1399 h&e-stained sentinel lymph node sections of breast cancer patients: the camelyon dataset

    Geert Litjens, Peter Bandi, Babak Ehteshami Bejnordi, Oscar Geessink, Maschenka Balkenhol, et al. 1399 h&e-stained sentinel lymph node sections of breast cancer patients: the camelyon dataset. GigaScience, 7 0 (6): 0 giy065, 2018

    work page 2018

  21. [21]

    Abnormality detection in chest X -ray images using uncertainty prediction autoencoders

    Yifan Mao, Fei-Fei Xue, Ruixuan Wang, Jianguo Zhang, Wei-Shi Zheng, and Hongmei Liu. Abnormality detection in chest X -ray images using uncertainty prediction autoencoders. In Medical Image Computing and Computer Assisted Intervention -- MICCAI 2020 , volume 12266 of Lecture Notes in Computer Science, pages 529--538. Springer, 2020

    work page 2020

  22. [22]

    Umap: Uniform manifold approximation and projection

    Leland McInnes, John Healy, Nathaniel Saul, and Lukas Grossberger. Umap: Uniform manifold approximation and projection. The Journal of Open Source Software, 3 0 (29): 0 861, 2018

    work page 2018

  23. [23]

    On the pitfalls of using the residual error as anomaly score

    Felix Meissen, Benedikt Wiestler, Georgios Kaissis, and Daniel Rueckert. On the pitfalls of using the residual error as anomaly score. In International Conference on Medical Imaging with Deep Learning, pages 914--928. PMLR, 2022

    work page 2022

  24. [24]

    Vindr-cxr: An open dataset of chest x-rays with radiologist's annotations

    Ha Q Nguyen, Khanh Lam, Linh T Le, Hieu H Pham, Dat Q Tran, Dung B Nguyen, Dung D Le, et al. Vindr-cxr: An open dataset of chest x-rays with radiologist's annotations. Scientific Data, 9 0 (1): 0 429, 2022

    work page 2022

  25. [25]

    Brain tumor mri dataset

    Masoud Nickparvar. Brain tumor mri dataset. https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset, 2021. Accessed: 2026-04-17

    work page 2021

  26. [26]

    Relative NN-Descent: A Fast Index Construction for Graph-Based Approximate Nearest Neighbor Search

    Naoki Ono and Yusuke Matsui. Relative nn-descent: A fast index construction for graph-based approximate nearest neighbor search. arXiv preprint arXiv:2310.20419, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  27. [27]

    Mean-shifted contrastive loss for anomaly detection

    Tal Reiss and Yedid Hoshen. Mean-shifted contrastive loss for anomaly detection. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 2155--2162, 2023

    work page 2023

  28. [28]

    Panda: Adapting pretrained features for anomaly detection and segmentation

    Tal Reiss, Niv Cohen, Liron Bergman, and Yedid Hoshen. Panda: Adapting pretrained features for anomaly detection and segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2806--2814, 2021

    work page 2021

  29. [29]

    Self-supervised predictive convolutional attentive block for anomaly detection

    Nicolae-C a t a lin Ristea, Neelu Madan, Radu Tudor Ionescu, Kamal Nasrollahi, Fahad Shahbaz Khan, Thomas B Moeslund, and Mubarak Shah. Self-supervised predictive convolutional attentive block for anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13576--13586, 2022

    work page 2022

  30. [30]

    Lukas et al. Ruff. Unifying deep one-class classification and anomaly detection. IEEE TPAMI, 2021

    work page 2021

  31. [31]

    Anatomy-aware self-supervised learning for anomaly detection in chest radiographs

    Junya Sato, Yuki Suzuki, Toshihiro Wataya, Daiki Nishigaki, Koji Kita, Kazuki Yamagata, Noriyuki Tomiyama, and Shoji Kido. Anatomy-aware self-supervised learning for anomaly detection in chest radiographs. Iscience, 26 0 (11), 2023

    work page 2023

  32. [32]

    Unsupervised anomaly detection with generative adversarial networks to guide marker discovery

    Thomas Schlegl, Philipp Seeb \"o ck, Sebastian M Waldstein, Ursula Schmidt-Erfurth, and Georg Langs. Unsupervised anomaly detection with generative adversarial networks to guide marker discovery. In International Conference on Information Processing in Medical Imaging, pages 146--157. Springer, 2017

    work page 2017

  33. [33]

    f-anogan: Fast unsupervised anomaly detection with generative adversarial networks

    Thomas Schlegl, Philipp Seeb \"o ck, Sebastian M Waldstein, Georg Langs, and Ursula Schmidt-Erfurth. f-anogan: Fast unsupervised anomaly detection with generative adversarial networks. Medical image analysis, 54: 0 30--44, 2019

    work page 2019

  34. [34]

    Nina Shvetsova, Bart Bakker, Irina Fedulova, Heinrich Schulz, and Dmitry V. Dylov. Anomaly detection in medical imaging with deep perceptual autoencoders. IEEE Access, 9: 0 118571--118583, 2021

    work page 2021

  35. [35]

    Learning and evaluating representations for deep one-class classification, 2021

    Kihyuk Sohn, Chun-Liang Li, Jinsung Yoon, Minho Jin, and Tomas Pfister. Learning and evaluating representations for deep one-class classification, 2021. ICLR 2021, arXiv:2103.01194

    work page arXiv 2021

  36. [36]

    Rsna pneumonia detection challenge, 2018

    Anouk Stein, Carol Wu, Chris Carr, George Shih, Jamie Dulkowski, et al. Rsna pneumonia detection challenge, 2018

    work page 2018

  37. [37]

    Constrained contrastive distribution learning for unsupervised anomaly detection and localisation in medical images

    Yu Tian, Guansong Pang, Fengbei Liu, Yuanhong Chen, Seon Ho Shin, Johan W Verjans, Rajvinder Singh, and Gustavo Carneiro. Constrained contrastive distribution learning for unsupervised anomaly detection and localisation in medical images. In Medical Image Computing and Computer Assisted Intervention--MICCAI 2021, pages 128--140. Springer, 2021

    work page 2021

  38. [38]

    Unsupervised anomaly localization using variational auto-encoders

    David Zimmerer, Fabian Isensee, Jens Petersen, Simon Kohl, and Klaus Maier-Hein. Unsupervised anomaly localization using variational auto-encoders. In Medical Image Computing and Computer Assisted Intervention--MICCAI 2019, pages 289--297. Springer, 2019

    work page 2019