No Room for Squares: Using Bitmap Masks to Improve Pedestrian Detection Using CNNS.

Adam Warde; Hamza Yous; David Gregg; David Moloney

doi:10.21427/9vge-x813

No Room for Squares: Using Bitmap Masks to Improve Pedestrian Detection Using CNNS.

Adam Warde, Hamza Yous, David Gregg, David Moloney

Research output: Contribution to conference › Paper › peer-review

Abstract

In this paper we investigate a method to reduce the number of computations and associated activations in Convolutional Neural Networks (CNN) by using bitmaps. The bitmaps are used to mask the input images to the network that fall within a rectangular window but do not fall within the boundaries of the objects the network is being trained upon. The mask has the effect of rendering the operations on these portions of the training images trivial. The thesis is that applying this approach to CNNs will not degrade accuracy while at the same time reducing the computational workload and reducing memory footprint. We found that we can remove up to 60% of the input images and see no decrease in accuracy. This leads to activation sparsity that can be exploited using a hardware accelerator to speedup training and inference, and decrease energy consumed.

Original language	English
DOIs	https://doi.org/10.21427/9vge-x813
Publication status	Published - 1 Jan 2019
Externally published	Yes
Event	IMVIP 2019: Irish Machine Vision & Image Processing - Technological University Dublin, Dublin, Ireland Duration: 28 Aug 2019 → 30 Aug 2019

Conference

Conference	IMVIP 2019: Irish Machine Vision & Image Processing
Country/Territory	Ireland
City	Dublin
Period	28/08/19 → 30/08/19

Keywords

bitmap masks
pedestrian detection
Convolutional Neural Networks
CNN
computational workload
memory footprint
activation sparsity
hardware accelerator
training
inference
energy consumption

Access to Document

10.21427/9vge-x813Licence: CC BY-NC-SA

Cite this

@conference{a3d53daf0abe4aa5996653079b65367a,

title = "No Room for Squares: Using Bitmap Masks to Improve Pedestrian Detection Using CNNS.",

abstract = "In this paper we investigate a method to reduce the number of computations and associated activations in Convolutional Neural Networks (CNN) by using bitmaps. The bitmaps are used to mask the input images to the network that fall within a rectangular window but do not fall within the boundaries of the objects the network is being trained upon. The mask has the effect of rendering the operations on these portions of the training images trivial. The thesis is that applying this approach to CNNs will not degrade accuracy while at the same time reducing the computational workload and reducing memory footprint. We found that we can remove up to 60% of the input images and see no decrease in accuracy. This leads to activation sparsity that can be exploited using a hardware accelerator to speedup training and inference, and decrease energy consumed.",

keywords = "bitmap masks, pedestrian detection, Convolutional Neural Networks, CNN, computational workload, memory footprint, activation sparsity, hardware accelerator, training, inference, energy consumption",

author = "Adam Warde and Hamza Yous and David Gregg and David Moloney",

year = "2019",

month = jan,

day = "1",

doi = "10.21427/9vge-x813",

language = "English",

note = "IMVIP 2019: Irish Machine Vision & Image Processing ; Conference date: 28-08-2019 Through 30-08-2019",

}

TY - CONF

T1 - No Room for Squares: Using Bitmap Masks to Improve Pedestrian Detection Using CNNS.

AU - Warde, Adam

AU - Yous, Hamza

AU - Gregg, David

AU - Moloney, David

PY - 2019/1/1

Y1 - 2019/1/1

N2 - In this paper we investigate a method to reduce the number of computations and associated activations in Convolutional Neural Networks (CNN) by using bitmaps. The bitmaps are used to mask the input images to the network that fall within a rectangular window but do not fall within the boundaries of the objects the network is being trained upon. The mask has the effect of rendering the operations on these portions of the training images trivial. The thesis is that applying this approach to CNNs will not degrade accuracy while at the same time reducing the computational workload and reducing memory footprint. We found that we can remove up to 60% of the input images and see no decrease in accuracy. This leads to activation sparsity that can be exploited using a hardware accelerator to speedup training and inference, and decrease energy consumed.

AB - In this paper we investigate a method to reduce the number of computations and associated activations in Convolutional Neural Networks (CNN) by using bitmaps. The bitmaps are used to mask the input images to the network that fall within a rectangular window but do not fall within the boundaries of the objects the network is being trained upon. The mask has the effect of rendering the operations on these portions of the training images trivial. The thesis is that applying this approach to CNNs will not degrade accuracy while at the same time reducing the computational workload and reducing memory footprint. We found that we can remove up to 60% of the input images and see no decrease in accuracy. This leads to activation sparsity that can be exploited using a hardware accelerator to speedup training and inference, and decrease energy consumed.

KW - bitmap masks

KW - pedestrian detection

KW - Convolutional Neural Networks

KW - CNN

KW - computational workload

KW - memory footprint

KW - activation sparsity

KW - hardware accelerator

KW - training

KW - inference

KW - energy consumption

U2 - 10.21427/9vge-x813

DO - 10.21427/9vge-x813

M3 - Paper

T2 - IMVIP 2019: Irish Machine Vision & Image Processing

Y2 - 28 August 2019 through 30 August 2019

ER -

No Room for Squares: Using Bitmap Masks to Improve Pedestrian Detection Using CNNS.

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this