#import "@preview/postercise:0.2.0": *
#import themes.boxes: *
#import "@preview/fletcher:0.5.8" as fletcher: diagram, edge, node

#set page(width: 16in, height: 22in)
#set text(size: 16pt)

#show: theme.with(
  primary-color: rgb(28, 55, 103), // Dark blue
  background-color: white,
  accent-color: rgb(243, 163, 30), // Yellow
  titletext-color: white,
  titletext-size: 1.8em,
)

#poster-header(
  title: [Can SAM "Segment Anything"?  #linebreak() ],
  subtitle: [Evaluating Zero-Shot Performance on Crack Detection],
  authors: [Hanwen Yu],
  affiliation: [School of Advanced Technology, Supervisor: SiYue Yu
  ],
  logo-2: image("./img/xjtlu-o.png", width: 15em),
)


// #image("examples.png", width: 100%)


#poster-content(col: 3)[
  // Content goes here
  #normal-box(color: none)[
    == Introduction
    he Segment Anything Model (SAM) has demonstrated remarkable
    zero-shot segmentation capabilities on natural images. However, its zero-shot performance on domain-specific tasks remains underexplored.
    // WHY CRACK SEGMENTATION?
    // • Critical for infrastructure safety monitoring
    // • Challenging characteristics:
    // - Thin, elongated structures (often 1-5 pixels wide)
    // - Low contrast against background
    // - Complex branching topology

    // RESEARCH QUESTION

    *Can SAM2 achieve competitive crack segmentation
    performance without domain-specific training?*
    // CONTRIBUTIONS
    // • First systematic evaluation of SAM2 zero-shot capability
    // on crack segmentation
    // • Comprehensive comparison of prompt strategies
    // (bounding box vs. point-based prompts)
    // • Analysis of failure modes and practical limitations

  ]
  #normal-box(color: none)[
    == Methodology


    *Dataset*

    - Crack500: 500 images with pixel-wise annotations

    - Test set: 100 images for evaluation

    *Prompt Strategies*

    We evaluate four prompt generation approaches:

    #table(
      columns: 2,

      [Prompt Type], [Description],
      [Bounding Box], [Tight box around ground truth mask],
      [1-Point Prompt], [Single point sampled from GT skeleton (morphological center)],
      [3-Point Prompt], [Three uniformly distributed points along GT skeleton],
      [5-Point Prompt], [Five uniformly distributed points along GT skeleton],
    )

    *Evaluation*

    $
      "IoU" = "TP" / ("TP" + "FP" + "FN")
    $

    $
      "F1" = 2 * ("Precision" * "Recall") / ("Precision" + "Recall")
    $

    *Baselines*

    Supervised models: UNet, DeepCrack, TransUNet,
    CT-CrackSeg, VM-UNet, CrackSegMamba


    #import fletcher.shapes: brace, diamond, hexagon, parallelogram, pill

    #set text(size: 16pt)
    #diagram(
      node-fill: gradient.radial(white, blue, radius: 200%),
      node-stroke: blue,
      spacing: 25pt,
      (
        node((0, 0), [Crack Image], shape: rect),
        node((0, 1), [SAM Image Encoder], shape: rect),
        node((0, 2), [Prompt Generation #linebreak() BBox, 1/3/5 points], shape: rect),
        node((1, 2), [SAM Mask Decoder], shape: rect),
        node((1, 1), [Predircted Mask], shape: rect),
        node((1, 0), [Metrics (IoU, F1)], shape: rect),
      )
        .intersperse(edge("-|>"))
        .join(),
    )

  ]
  #normal-box(color: none)[
    == Experiments and Results
    #image("img/examples.png")
    #image("img/metrics.png")
    #image("img/sam_iou.png")
    #image("img/sam_f1.png")

  ]

  #normal-box(color: none)[

    == Qualitative Analysis
    #image("img/fail1.png")
    #image("img/fail2.png")

  ]

  #normal-box(color: none)[
    == Key Findings and Discussion
    // *Prompt Effectiveness*

    Bounding box prompts yield the best performance among zero-shot methods. There is a 4.7x performance gap between bbox(39.6% IoU) and 1-point prompts(8.4% IoU).


    SAM2 with bbox prompts (39.6% IoU) lags behind supervised models, even UNet in 2015. which highlights limitations of zero-shot approach without fine-tuning.

    // *Single Point Prompt Limitations*
    1-point prompts perform poorly (12.3% IoU), indicating insufficient guidance for complex crack structures.    5-point prompts approach bbox performance for highly irregular cracks, suggesting multiple points help capture shape.
  ]

  #normal-box(color: none)[

    == Conclusion and Future Work

    SAM2 shows limited zero-shot capability for crack segmentation. Bounding box prompts significantly outperform point-based prompts. Performance still lags behind supervised methods, indicating need for domain adaptation.
  ]
  #poster-footer[
    // Content
    Hanwen Yu | Email: Hanwen.Yu24\@student.xjtlu.edu.cn
  ]
]