Evaluation of Group Convolution in Lightweight Deep Networks for Object Classification

Das, Apurba; Boulay, Thomas; Yogamani, Senthil; Ou, Shiping

doi:10.1007/978-3-030-12177-8_5

Cited by 2 publications

(3 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Indeed, the layer sequence of our network is mainly inspired by the plain network of Resnet, no shortcut connections are used for our network which come up to the VGG architecture. The motivation for not using the shortcut connections comes from previous experiments which are detailed in this paper [3]. The authors demonstrate that the shortcut connections don't provide a significant improvement in accuracy in shallow networks, which is the case with our 9 layers encoder.…”

Section: Encoder Stage : V2n9slimmentioning

confidence: 98%

YUVMultiNet: Real-time YUV multi-task CNN for autonomous driving

Boulay,

El-Hachimi,

Surisetti

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

In this paper, we propose a multi-task convolutional neural network (CNN) architecture optimized for a low power automotive grade SoC. We introduce a network based on a unified architecture where the encoder is shared among the two tasks namely detection and segmentation. The proposed network runs at 25 FPS for 1280 × 800 resolution. We briefly discuss the methods used to optimize the network architecture such as using native YUV image directly, optimization of layers & feature maps and applying quantization. We also focus on memory bandwidth in our design as convolutions are data intensives and most SOCs are bandwidth bottlenecked. We then demonstrate the efficiency of our proposed network for a dedicated CNN accelerators presenting the key performance indicators (KPI) for the detection and segmentation tasks obtained from the hardware execution and the corresponding run-time.

show abstract

Section: Encoder Stage : V2n9slimmentioning

confidence: 98%

YUVMultiNet: Real-time YUV multi-task CNN for autonomous driving

Boulay,

El-Hachimi,

Surisetti

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…The current work extends this concept by adding group convolution in all convolution layers (Net-2, Net-3, Net-4, SoildNet), also we experiment with dynamic group size (Net-4, SoildNet) to reduce the network complexity by more than two times in trainable parameters (Net-3 vs. Net-4). The network schemes do not contain residual connections because group convolution was found to be not very effective for the networks with low depth as presented in [15]. Also, a similar study on residual connection for lightweight networks [16] is the reason not to use them in any of the proposals.…”

Section: Group Convolutionmentioning

confidence: 99%

“…The network schemes do not contain residual connections because group convolution was found to be not very effective for the networks with low depth as presented in [15]. Also, a similar study on residual connection for lightweight networks [16] is the reason not to use them in any of the proposals. While adding group convolution at all layers of the network brings another challenge of insufficient feature blending.…”

Section: Group Convolutionmentioning

confidence: 99%

SoildNet: Soiling Degradation Detection in Autonomous Driving

Das

2019

Preprint

Self Cite

View full text Add to dashboard Cite

In the field of autonomous driving, camera sensors are extremely prone to soiling because they are located outside of the car and interact with environmental sources of soiling such as rain drops, snow, dust, sand, mud and so on. This can lead to either partial or complete vision degradation. Hence detecting such decay in vision is very important for safety and overall to preserve the functionality of the "autonomous" components in autonomous driving. The contribution of this work involves: 1) Designing a Deep Convolutional Neural Network (DCNN) based baseline network, 2) Exploiting several network remodelling techniques such as employing static and dynamic group convolution, channel reordering to compress the baseline architecture and make it suitable for low power embedded systems with ∼1 TOPS, 3) Comparing various result metrics of all interim networks dedicated for soiling degradation detection at tile level of size 64 × 64 on input resolution 1280 × 768. The compressed network, is called SoildNet (Sand, snOw, raIn/dIrt, oiL, Dust/muD) that uses only 9.72% trainable parameters of the base network and reduces the model size by more than 7 times with no loss in accuracy.

show abstract

Evaluation of Group Convolution in Lightweight Deep Networks for Object Classification

Cited by 2 publications

References 13 publications

YUVMultiNet: Real-time YUV multi-task CNN for autonomous driving

YUVMultiNet: Real-time YUV multi-task CNN for autonomous driving

SoildNet: Soiling Degradation Detection in Autonomous Driving

Contact Info

Product

Resources

About