Deep Semantic Instance Segmentation of Tree-Like Structures Using Synthetic Data

Halupka, Kerry; Garnavi, Rahil; Moore, Stephen

doi:10.1109/wacv.2019.00187

Cited by 6 publications

(3 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We performed all experiments with PyTorch. 8 We trained our nets on four NVIDIA Titan X Pascal GPUs using synchronous gradient update. We used the AMSGrad variant [28] of the Adam optimizer [14], with α = 0.001, β 1 = 0.9, β 2 = 0.999, and = 10 −8 .…”

Section: Methodsmentioning

confidence: 99%

“…We use the discriminative loss function proposed by Brabandere et al [3] to learn dense voxel embeddings. This loss function has recently been adopted by Luther & Seung [22] in the application of 2D neuron segmentation, and has also been applied to an increasingly wide variety of other problem domains including object discovery in videos [34], 3D point cloud embeddings [27,5,19], scene text detection [30], biomedical segmentation [8], driver information systems [21], 3D volumetric instance segmentation [16,17], single-image piece-wise planar 3D Reconstruction [35].…”

Section: Means-based Loss Functionmentioning

confidence: 99%

See 1 more Smart Citation

Learning and Segmenting Dense Voxel Embeddings for 3D Neuron Reconstruction

Lee¹,

Lu²,

Luther³

et al. 2019

Preprint

View full text Add to dashboard Cite

We show dense voxel embeddings learned via deep metric learning can be employed to produce a highly accurate segmentation of neurons from 3D electron microscopy images. A metric graph on an arbitrary set of short and long-range edges can be constructed from the dense embeddings generated by a convolutional network. Partitioning the metric graph with long-range affinities as repulsive constraints can produce an initial segmentation with high precision, with substantial improvements on very thin objects. The convolutional embedding net is reused without any modification to agglomerate the systematic splits caused by complex "self-touching" objects. Our proposed method achieves state-ofthe-art accuracy on the challenging problem of 3D neuron reconstruction from the brain images acquired by serial section electron microscopy. Our alternative, object-centered representation could be more generally useful for other computational tasks in automated neural circuit reconstruction.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Means-based Loss Functionmentioning

confidence: 99%

Learning and Segmenting Dense Voxel Embeddings for 3D Neuron Reconstruction

Lee¹,

Lu²,

Luther³

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…By now, no predecessor has ever combined object detection, semantic segmentation and linear fitting in antenna parameter measurement. For the first time, this paper validated a fully automatic antenna parameter measurement method based on instance segmentation [29][30][31], least squares, frame sequence analysis and UAV [32][33][34][35], which enjoys remarkable preciseness, rapid recognition and outstanding performance.…”

Section: Related Workmentioning

confidence: 99%

Mobile Communication Base Station Antenna Measurement Using Unmanned Aerial Vehicle

Zhai

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Traditional base station antenna measurement methods conducted with professional worker climbing towers tend to raise safety and inefficiency concerns in practical application. Designed to address the above problems, this paper proposes an intelligent and fully automatic antenna measurement unmanned aerial vehicle (UAV) system for mobile communication base station. Firstly, an antenna database, containing 19,715 images, named UAV-Antenna is constructed by image capturing with the help of UAVs flying around various base stations. Secondly, Mask R-CNN is adopted to train an optimal instance segmentation model on UAV-Antenna. Then, pixel coordinates and threshold are utilized for measuring antenna quantity and separate all antenna data for further measuring. Finally, a least squares method is employed for measuring antenna parameters. Experimental results show that the proposed method can not only satisfy the industry application standards, but also guarantee safety of labors and efficiency of performance.

show abstract

Use of Individual Tree and Product Level Data to Improve Operational Forestry

Keefe

Zimbelman

Picchi³

2022

Curr Forestry Rep

View full text Add to dashboard Cite

Purpose of Review Individual tree detection (ITD) methods and technologies for tracking individual forest products through a smart operational supply chain from stump to mill are now available. The purpose of this paper is to (1) review the related literature for audiences not familiar with remote sensing and tracking technologies and (2) to identify knowledge gaps in operational forestry and forest operations research now that these new data and systems are becoming more common. Recent Findings Past research has led to successful development of ITD remote sensing methods for detecting individual tree information and radio frequency identification (RFID), branding, and other product tracing methods for individual trees and logs. Blockchain and cryptocurrency that allow independent verification of transactions and work activity recognition based on mobile and wearable sensors can connect the mechanized and motor-manual components of supply chains, bridging gaps in the connectivity of data. However, there is a shortage of research demonstrating use of location-aware tree and product information that spans multiple machines. Summary Commercial products and technologies are now available to digitalize forest operations. Research should shift to evaluation of applications that demonstrate use. Areas for improved efficiencies include (1) use of wearable technology to map individual seedlings during planting; (2) optimizing harvesting, skidding and forwarder trails, landings, and decking based on prior knowledge of tree and product information; (3) incorporation of high-resolution, mapped forest product value and treatment cost into harvest planning; (4) improved machine navigation, automation, and robotics based on prior knowledge of stem locations; (5) use of digitalized silvicultural treatments, including microclimate-smart best management practices; and (6) networking of product tracking across multiple, sensorized machines.

show abstract

Deep Semantic Instance Segmentation of Tree-Like Structures Using Synthetic Data

Cited by 6 publications

References 30 publications

Learning and Segmenting Dense Voxel Embeddings for 3D Neuron Reconstruction

Learning and Segmenting Dense Voxel Embeddings for 3D Neuron Reconstruction

Mobile Communication Base Station Antenna Measurement Using Unmanned Aerial Vehicle

Use of Individual Tree and Product Level Data to Improve Operational Forestry

Contact Info

Product

Resources

About