Detect Anything in Real Time: From Single-Prompt Segmentation to Multi-Class Detection

Dimensions overall score 9.0

GitHub Code Pulse

No public code linked for this paper yet.

• DART achieves a 5.6x cumulative speedup at 3 classes and scales to 25x at 80 classes compared to SAM3's single-prompt processing.result 95%
  Evidencepartial

  these optimizations yield 5.6x cumulative speedup at 3 classes, scaling to 25x at 80 classes

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• DART reduces the computational cost of the visual backbone from O(N) to O(1) by sharing backbone computation across all classes.technical 95%
  Evidencepartial

  This allows the backbone computation to be shared between all classes, reducing its cost from O(N) to O(1)

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• On COCO val2017, DART achieves 55.8 AP at 15.8 FPS (4 classes, 1008x1008) on a single RTX 4080.result 95%
  Evidencepartial

  On COCO val2017 (5,000 images, 80 classes), DART achieves 55.8 AP at 15.8 FPS (4 classes, 1008x1008) on a single RTX 4080

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• DART surpasses purpose-built open-vocabulary detectors trained on millions of box annotations in terms of performance.result 90%
  Evidencepartial

  surpassing purpose-built open-vocabulary detectors trained on millions of box annotations

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• SAM3 processes only a single text prompt per forward pass, requiring N independent executions for N categories.limitation 95%
  Evidencepartial

  SAM3 processes a single text prompt per forward pass. Detecting N categories requires N independent executions

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• DART is a training-free framework that converts SAM3 into a real-time multi-class detector without modifying any model weight.method 95%
  Evidencepartial

  We present Detect Anything in Real Time (DART), a training-free framework that converts SAM3 into a real-time multi-class detector

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• For extreme latency targets, adapter distillation with a frozen encoder-decoder achieves 38.7 AP with a 13.9 ms backbone.result 90%
  Evidencepartial

  For extreme latency targets, adapter distillation with a frozen encoder-decoder achieves 38.7 AP with a 13.9 ms backbone

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• DART exploits the structural invariant that SAM3's visual backbone is class-agnostic, producing image features independent of the text prompt.technical 90%
  Evidencepartial

  by exploiting a structural invariant: the visual backbone is class-agnostic, producing image features independent of the text prompt

  Implicationmissing

  Implication not extracted yet.

  Verificationpartial

  partial

• DART converts SAM3 into a real-time multi-class detector without any training or weight modification.method 95%
  Evidencepartial

  We present Detect Anything in Real Time (DART), a training-free framework that converts SAM3 into a real-time multi-class detector by exploiting a structural invariant

  Implicationpartial

  Directly stated in abstract: 'training-free framework' and 'without modifying any model weight'.

  Verificationpartial

  partial

• DART reduces the backbone computation cost from O(N) to O(1) by sharing visual features across all classes.technical 95%
  Evidencepartial

  the visual backbone is class-agnostic, producing image features independent of the text prompt. This allows the backbone computation to be shared between all classes, reducing its cost from O(N) to O(1)

  Implicationpartial

  Explicitly stated in abstract with complexity notation.

  Verificationpartial

  partial

• DART achieves 5.6x cumulative speedup at 3 classes and 25x at 80 classes compared to SAM3.result 95%
  Evidencepartial

  these optimizations yield 5.6x cumulative speedup at 3 classes, scaling to 25x at 80 classes

  Implicationpartial

  Directly stated with specific numbers in abstract.

  Verificationpartial

  partial

• On COCO val2017 with 4 classes and 1008x1008 resolution, DART achieves 55.8 AP at 15.8 FPS on a single RTX 4080.result 95%
  Evidencepartial

  On COCO val2017 (5,000 images, 80 classes), DART achieves 55.8 AP at 15.8 FPS (4 classes, 1008x1008) on a single RTX 4080

  Implicationpartial

  Explicit numeric result stated in abstract.

  Verificationpartial

  partial

Startup potential card