# MiLSD: A Micro Line-Segment Detector for Resource-Constrained Devices

arXiv: https://arxiv.org/abs/2607.06600 · HTML: https://mlsanity.com/paper/2607.06600/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 25 of its 40 references that are in the map.
- First posted on arXiv 7 Jul 2026.

## Builds on (most cited first, 25 of 25)

- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference](https://mlsanity.com/paper/1712.05877/) (2017)
- [Improved Knowledge Distillation via Teacher Assistant](https://mlsanity.com/paper/1902.03393/) (2019)
- [MCUNet: Tiny Deep Learning on IoT Devices](https://mlsanity.com/paper/2007.10319/) (2020)
- [CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs](https://mlsanity.com/paper/1801.06601/) (2018)
- [MCUNetV2: Memory-Efficient Patch-based Inference for Tiny Deep Learning](https://mlsanity.com/paper/2110.15352/) (2021)
- [Line Segment Detection Using Transformers without Edges](https://mlsanity.com/paper/2101.01909/) (2021)
- [End-to-End Wireframe Parsing](https://mlsanity.com/paper/1905.03246/) (2019)
- [Learning to Parse Wireframes in Images of Man-Made Environments](https://mlsanity.com/paper/2007.07527/) (2020)
- [Holistically-Attracted Wireframe Parsing](https://mlsanity.com/paper/2003.01663/) (2020)
- [DeepLSD: Line Segment Detection and Refinement with Deep Image Gradients](https://mlsanity.com/paper/2212.07766/) (2022)
- [Learning Attraction Field Representation for Robust Line Segment Detection](https://mlsanity.com/paper/1812.02122/) (2018)
- [Deep Hough Transform for Semantic Line Detection](https://mlsanity.com/paper/2003.04676/) (2020)
- [PPGNet: Learning Point-Pair Graph for Line Segment Detection](https://mlsanity.com/paper/1905.03415/) (2019)
- [Memory-Driven Mixed Low Precision Quantization For Enabling Deep Network Inference On Microcontrollers](https://mlsanity.com/paper/1905.13082/) (2019)
- [Towards Light-weight and Real-time Line Segment Detection](https://mlsanity.com/paper/2106.00186/) (2021)
- [TP-LSD: Tri-Points Based Line Segment Detector](https://mlsanity.com/paper/2009.05505/) (2020)
- [Holistically-Attracted Wireframe Parsing: From Supervised to Self-Supervised Learning](https://mlsanity.com/paper/2210.12971/) (2022)
- [Fully Convolutional Line Parsing](https://mlsanity.com/paper/2104.11207/) (2021)
- [ELSED: Enhanced Line SEgment Drawing](https://mlsanity.com/paper/2108.03144/) (2021)
- [Quantization and Deployment of Deep Neural Networks on Microcontrollers](https://mlsanity.com/paper/2105.13331/) (2021)
- [ULSD: Unified Line Segment Detection across Pinhole, Fisheye, and Spherical Cameras](https://mlsanity.com/paper/2011.03174/) (2020)
- [LSDNet: Trainable Modification of LSD Algorithm for Real-Time Line Segment Detection](https://mlsanity.com/paper/2209.04642/) (2022)
- [DT-LSD: Deformable Transformer-based Line Segment Detection](https://mlsanity.com/paper/2411.13005/) (2024)
- [LINEA: Fast and Accurate Line Detection Using Scalable Transformers](https://mlsanity.com/paper/2505.16264/) (2025)

## Built on by (most cited first, 1 of 1)

- [SweepLSD: A One-Pass, O(width)-Memory Line Segment Detector with an Integer-Only Streaming Core and a Real-Time FPGA Realization](https://mlsanity.com/paper/2608.22086/) (2026)

## Abstract (arXiv)

Line segment detection is a key building block in visual SLAM, 3D reconstruction, and industrial inspection. Recent deep learning methods have greatly improved accuracy, yet even the smallest models require several megabytes of memory, exceeding low-cost MCU capacity. This work investigates the maximum achievable accuracy under a sub-megabyte budget. We propose MiLSD, a detector tailored for MCU-level constraints, and systematically compare three output representations within a compact fully-convolutional backbone. Our study shows that the proposed F-Clip center-with-length-and-angle formulation learns most effectively at small model sizes. We find that 8-bit quantization preserves full-precision performance, while 4-bit quantization causes significant degradation, particularly in angle regression, with quantization-aware training recovering only part of the loss. With a one-megabyte activation budget and inference enhancements including sub-pixel decoding, test-time augmentation, and a lightweight verifier, MiLSD improves sAP10 on ShanghaiTech Wireframe from 10.6 (25k parameters, 0.25 MB) to 24.1 within 1 MB. Rather than competing with GPU-scale parsers, we map the accuracy memory trade-off across representations, bit-widths, capacities, and post-processing strategies for embedded vision systems.