Assisted Excitation of Activations: A Learning Technique to Improve Object Detectors

Derakhshani, Mohammad Mahdi; Masoudnia, Saeed; Shaker, Amir Hossein; Mersa, Omid; Sadeghi, Maryam; Rastegari, Mohammad; Araabi, Babak Nadjar

doi:10.1109/cvpr.2019.00942

Cited by 18 publications

(8 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Attention mechanism, a) considering (5–7) the attended feature map is obtained by applying the excitation factor ( α ) in a decreasing rate. The overall structure is similar to the work presented by Derakhshani et al[47] except for the W f and the adaptive excitation factor, b) adaptive excitation factor profile. Using this technique, the excitation factor is reduced when the metric does not improve on validation.…”

Section: Methodsmentioning

confidence: 81%

“…2.a represents a schematic of the AE module. We proposed an adaptive version of the excitation factor instead of a fixed monotonic function [47]. Rather than employing a smooth, monotonically decreasing function (8) to alter the attention coefficient, we degraded the parameter stepwise, after a few epochs with no enhancement in validation accuracy.…”

Section: Methodsmentioning

confidence: 99%

“…Assisted excitation (AE) of activation is a strategy to enhance recall by employing ground truth bounding boxes in the training process. The proposed attention in the AE structure was adapted from Derakhshani et al [47] and is shown in Figure 2a. The excitation was done by applying the ground truth bounding boxes on the filtered tensor.…”

Section: B Attention Mechanismmentioning

confidence: 99%

“…Some studies take a different approach to solve the problem. Derakhshani et al [47] used ground truth bounding boxes to excite more effective activations during training. The activations in the vicinity of the objects are excited more at the beginning of the training and gradually decreased to zero until the end of training.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Spatiotemporal analysis of speckle dynamics to track invisible needle in ultrasound sequences using Convolutional Neural Networks

Zade

Aziz

Majedi

et al. 2022

Preprint

View full text Add to dashboard Cite

ObjectiveAccurate needle placement to the target point is critical for ultrasound interventions like biopsies and epidural injections. However, aligning the needle to the thin plane of the transducer is a challenging issue as it leads to the decay of visibility by the naked eye. Therefore, we have developed a CNN-based framework to track the needle using the spatiotemporal features of speckle dynamics.MethodsThere are three key techniques to optimize the network for our application. First, we proposed a motion field estimation network (RMF) to extract spatiotemporal features from the stack of consecutive frames. We also designed an efficient network based on the state-of-the-art Yolo framework (nYolo). Lastly, the Assisted Excitation (AE) module was added at the neck of the network to handle imbalance problem.ResultsTen freehand ultrasound sequences are collected by inserting an injection needle steeply into the Ultrasound Compatible Lumbar Epidural Simulator and Femoral Vascular Access Ezono test phantoms. We divided the dataset into two sub-categories. In the second category, in which the situation is more challenging and the needle is totally invisible statically, the angle and tip localization error were 2.43±1.14° and 2.3±1.76 mm using Yolov3+RMF+AE and 2.08±1.18° and 2.12±1.43 mm using nYolo+RMF+AE.Conclusion and significanceThe proposed method has the potential to track the needle in a more reliable operation compared to other state-of-the-art methods and can accurately localize it in 2D B-mode US images in real-time, allowing it to be used in in current ultrasound intervention procedures.

show abstract

Section: Methodsmentioning

confidence: 81%

Section: Methodsmentioning

confidence: 99%

Section: B Attention Mechanismmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Spatiotemporal analysis of speckle dynamics to track invisible needle in ultrasound sequences using Convolutional Neural Networks

Zade

Aziz

Majedi

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…First, our proposal offers a compact model, with a lower number of layers in the network and smaller filters, and the most important point is that we use a specific type of layer that is not present in the original YOLO version using a technique called "Assisted Excitation of Activations" [32]. We also do not use pre-trained weights, as there is no model trained with the characteristics of ours, so we trained our model from the beginning.…”

Section: Implementation Detailsmentioning

confidence: 99%

Hpnet: Detecting Human Parts in the Wild / Hpnet: Detectando Partes Humanas Na Natureza

Casa

Souza

Oliveira

2021

BJD

View full text Add to dashboard Cite

The objective of this work is to present a model of neural network for the detection and segmentation of human's body parts (HPNet). We offer a multiplatform real-time solution that can be run on ordinary computers, as well as on mobile devices and embedded systems. Our proposal is characterized by presenting a compact solution, and by investigating a part of object detection still little explored. One of the striking features presented is the ability to recognize parts of the human body even in uncontrolled environments, due to the use of a random subset of Google's public database (Open Images Dataset) that contains images with objects in the most varied sizes, positions, lighting and occlusion conditions. At first, we offer a solution only for the detection and segmentation of the common parts of the human body, but we intend to expand its capabilities to detect other more specific parts and regions. The main purpose of our model is its use to solve specific problems that require the detection and segmentation of human's body parts, for example, in user authentication.

show abstract

Machine Learning Agricultural Application Based on the Secure Edge Computing Platform

Liu

et al. 2020

Machine Learning for Cyber Security

View full text Add to dashboard Cite

Assisted Excitation of Activations: A Learning Technique to Improve Object Detectors

Abstract: Our technique improves the mAP of YOLOv2 by 3.8% and mAP of YOLOv3 by 2.2% on MSCOCO dataset.This technique is inspired from curriculum learning. It is simple and effective and it is applicable to most single-stage object detectors.

Cited by 18 publications

References 32 publications

Spatiotemporal analysis of speckle dynamics to track invisible needle in ultrasound sequences using Convolutional Neural Networks

Spatiotemporal analysis of speckle dynamics to track invisible needle in ultrasound sequences using Convolutional Neural Networks

Hpnet: Detecting Human Parts in the Wild / Hpnet: Detectando Partes Humanas Na Natureza

Machine Learning Agricultural Application Based on the Secure Edge Computing Platform

Contact Info

Product

Resources

About