paper-with-me

홈 › Papers

When Tracking Fails: Analyzing Failure Modes of SAM2 for Point-Based Tracking in Surgical Videos

2025-10-02 · Woowon Jang, Jiwon Im, Juseung Choi, Niki Rashidian, Wesley De Neve, Utku Ozbulak arxiv

Video object segmentation (VOS) models such as SAM2 offer promising zero-shot tracking capabilities for surgical videos using minimal user input. Among the available input types, point-based tracking offers an efficient and low-cost alternative, yet its reliability and failure cases in complex surgical environments are not well understood. In this work, we systematically analyze the failure modes of point-based tracking in laparoscopic cholecystectomy videos. Focusing on three surgical targets, the gallbladder, grasper, and L-hook electrocautery, we compare the performance of point-based tracking with segmentation mask initialization. Our results show that point-based tracking is competitive for surgical tools but consistently underperforms for anatomical targets, where tissue similarity and ambiguous boundaries lead to failure. Through qualitative analysis, we reveal key factors influencing tracking outcomes and provide several actionable recommendations for selecting and placing tracking points to improve performance in surgical video analysis.

📄 PDF Abstract BibTeX arXiv:2510.02100

Code (0)

등록된 구현이 없습니다.

Tasks

Video Object Segmentation

Similar Papers 제목 키워드 기반

Multiparty Dynamics and Failure Modes for Machine Learning and Artificial Intelligence

2018-10-16 · David Manheim

An important challenge for safety in machine learning and artificial intelligence systems is a~set of related failures involving specification gaming, reward hacking, fragility to distributional shifts, and Goodhart's or…

BIG-bench Machine Learning

Jailbroken: How Does LLM Safety Training Fail?

2023-07-05 · NeurIPS 2023 11

Large language models trained for safety and harmlessness remain susceptible to adversarial misuse, as evidenced by the prevalence of "jailbreak" attacks on early releases of ChatGPT that elicit undesired behavior. Going…

Red Teaming

AI Alignment Strategies from a Risk Perspective: Independent Safety Mechanisms or Shared Failures?

2025-10-13 · Leonard Dung, Florian Mai arxiv

AI alignment research aims to develop techniques to ensure that AI systems do not cause harm. However, every alignment technique has failure modes, which are conditions in which there is a non-negligible chance that the …

ProPINN: Demystifying Propagation Failures in Physics-Informed Neural Networks

2025-02-02 · Haixu Wu, Yuezhou Ma, Hang Zhou, Huikun Weng 외

Physics-informed neural networks (PINNs) have earned high expectations in solving partial differential equations (PDEs), but their optimization usually faces thorny challenges due to the unique derivative-dependent loss …

Select-And-Extract: A Lightweight Plugin for Retrieval-Augmented Generation

2026-08-01 · Chenming Tang, Jiawei Han arxiv

Retrieval-augmented generation (RAG) for language model (LM) systems fundamentally has two failure modes: retrieval failure and reading failure. The former fails to recall the right pieces of information from the externa…