Filters








80 Hits in 2.3 sec

DEM Super-Resolution with EfficientNetV2 [article]

Bekir Z Demiray, Muhammed Sit, Ibrahim Demir
2021 arXiv   pre-print
Efficient climate change monitoring and modeling rely on high-quality geospatial and environmental datasets.  ...  Digital Elevation Model (DEM) datasets are such examples whereas their low-resolution versions are widely available, high-resolution ones are scarce.  ...  In the proposed model, we modify the MobileNetV3 blocks to use for super-resolution tasks.  ... 
arXiv:2109.09661v1 fatcat:lufkvcccmvgq5od2dio2aa4iry

SqueezeNAS: Fast neural architecture search for faster semantic segmentation [article]

Albert Shaw, Daniel Hunter, Forrest Iandola, Sammy Sidhu
2019 arXiv   pre-print
While Neural Architecture Search (NAS) has been effectively used to develop low-latency networks for image classification, there has been relatively little effort to use NAS to optimize DNN architectures  ...  We also compare our networks to the efficient segmentation networks proposed in MobileNetV3 [36] .  ...  Sampling from the Gumbel-Softmax distribution allows us to efficiently optimize the architecture distribution by using gradient descent on the stochastic supernetwork.  ... 
arXiv:1908.01748v2 fatcat:knlqepbolrdejigqwmnda5pzwq

Multi-path Neural Networks for On-device Multi-domain Visual Classification [article]

Qifei Wang, Junjie Ke, Joshua Greaves, Grace Chu, Gabriel Bender, Luciano Sbaiz, Alec Go, Andrew Howard, Feng Yang, Ming-Hsuan Yang, Jeff Gilbert, Peyman Milanfar
2021 arXiv   pre-print
MobileNetV3-like search space.  ...  MobileNetV3-like architectures.  ...  Based on the singledomain efficient NAS framework [1] , the proposed multipath NAS for MDL uses multiple reinforcement learning (RL) controllers, where each selects an optimal path from the super-network  ... 
arXiv:2010.04904v2 fatcat:x4ietulpqrcydhfumtxisy5pra

MicroNet: Improving Image Recognition with Extremely Low FLOPs [article]

Yunsheng Li and Yinpeng Chen and Xiyang Dai and Dongdong Chen and Mengchen Liu and Lu Yuan and Zicheng Liu and Lei Zhang and Nuno Vasconcelos
2021 arXiv   pre-print
For instance, under the constraint of 12M FLOPs, MicroNet achieves 59.4\% top-1 accuracy on ImageNet classification, outperforming MobileNetV3 by 9.6\%.  ...  [44] adapts image resolution to achieve efficient inference. Another line of work keeps the architectures fixed, but adapts parameters.  ...  Note that input resolution 224×224 is used for MicroNet and related works other than HBONet/TinyNet, whose input resolution is shown in the bracket.  ... 
arXiv:2108.05894v1 fatcat:ablts26dijbfznzippauc2vioa

AIM 2020 Challenge on Efficient Super-Resolution: Methods and Results [article]

Kai Zhang, Martin Danelljan, Yawei Li, Radu Timofte, Jie Liu, Jie Tang, Gangshan Wu, Yu Zhu, Xiangyu He, Wenjie Xu, Chenghua Li, Cong Leng (+73 others)
2020 arXiv   pre-print
They gauge the state-of-the-art in efficient single image super-resolution.  ...  This paper reviews the AIM 2020 challenge on efficient single image super-resolution with focus on the proposed solutions and results.  ...  The core of this approach is to use modified MobileNetV3 [16] blocks to design an efficient method for SR.  ... 
arXiv:2009.06943v1 fatcat:2s7k5wsgsjgo5flnqaby26cn64

Model Rubik's Cube: Twisting Resolution, Depth and Width for TinyNets [article]

Kai Han, Yunhe Wang, Qiulin Zhang, Wei Zhang, Chunjing Xu, Tong Zhang
2020 arXiv   pre-print
The giant formula for simultaneously enlarging the resolution, depth and width provides us a Rubik's cube for neural networks.  ...  So that we can find networks with high efficiency and excellent performance by twisting the three dimensions.  ...  The straightforward way for designing tiny networks is to apply the experience used in Efficient-Net [44] .  ... 
arXiv:2010.14819v2 fatcat:v3rnohc26bee7o6lfbg3t3ytwa

Factorizing and Reconstituting Large-Kernel MBConv for Lightweight Face Recognition

Yaqi Lyu, Jing Jiang, Kun Zhang, Yilun Hua, Miao Cheng
2019 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)  
We combine FR-MBConv with MobileNetV3 [16] to build a lightweight face recognition model.  ...  NAS normally use hand-craft MBConv as building block. However, they mainly searched for block-related hyperparameters, and the structure of MBConv itself was largely overlooked.  ...  In this section, based on a SOTA hardware-aware efficient MobileNetV3-large, we built a lightweight face recognition model.  ... 
doi:10.1109/iccvw.2019.00329 dblp:conf/iccvw/LyuJZHC19 fatcat:zodpnifw5vbxtd6hscbtxigf3a

Towards Efficient and Data Agnostic Image Classification Training Pipeline for Embedded Systems [article]

Kirill Prokofiev, Vladislav Sovrasov
2021 arXiv   pre-print
Resulting models are computationally efficient and can be deployed to CPU using the OpenVINO toolkit.  ...  Nowadays deep learning-based methods have achieved a remarkable progress at the image classification task among a wide range of commonly used datasets (ImageNet, CIFAR, SVHN, Caltech 101, SUN397, etc.)  ...  Smith, L.N., Topin, N.: Super-convergence: very fast training of neural networks using large learning rates. In: Defense + Commercial Sensing (2019) 34.  ... 
arXiv:2108.07049v1 fatcat:c3i2mav6g5cr7p2rprtnqzhkfa

Application of Ghost-DeblurGAN to Fiducial Marker Detection [article]

Yibo Liu, Amaldev Haridevan, Hunter Schofield, Jinjun Shan
2022 arXiv   pre-print
The datasets and codes used in this paper are available at: https://github.com/York-SDCNLab/Ghost-DeblurGAN.  ...  Compared with BN which is not recommended in low-level tasks, such as super-resolution [37] , IN preserves more scale information and maintains the same normalization procedure in training and inference  ...  However, degradation occurs in terms of PSNR when using MobileNetV3 as the backbone, while the deblurring quality of GhostNet is comparable with that of MobileNetV2.  ... 
arXiv:2109.03379v3 fatcat:7tsa5qcgnzfalj2pwugju5eksy

Network Augmentation for Tiny Deep Learning [article]

Han Cai, Chuang Gan, Ji Lin, Song Han
2022 arXiv   pre-print
At test time, only the tiny model is used for inference, incurring zero inference overhead. We demonstrate the effectiveness of NetAug on image classification and object detection.  ...  Figure 5 demonstrates the results of YoloV3+MobileNetV2 w0.35 and YoloV3+MobileNetV3 w0.35 under different input resolutions.  ...  Its goal is to provide efficient performance estimation in NAS.  ... 
arXiv:2110.08890v2 fatcat:l5pcfamu6zghnl7loxlbyjyzbm

Fast Neural Architecture Search for Lightweight Dense Prediction Networks [article]

Lam Huynh, Esa Rahtu, Jiri Matas, Janne Heikkila
2022 arXiv   pre-print
The performance of LPD is evaluated on monocular depth estimation, semantic segmentation, and image super-resolution tasks on diverse datasets, including NYU-Depth-v2, KITTI, Cityscapes, COCO-stuff, DIV2K  ...  Starting from a pre-defined generic backbone, LDP applies the novel Assisted Tabu Search for efficient architecture exploration.  ...  Image super-resolution task has also been immensely improved using deep neural networks. Dong et al.  ... 
arXiv:2203.01994v3 fatcat:nnz34pody5banfrqpkaanpszau

Plant Leaf Disease Recognition Using Depth-Wise Separable Convolution-Based Models

Syed Mohammad Minhaz Hossain, Kaushik Deb, Pranab Kumar Dhar, Takeshi Koshiba
2021 Symmetry  
Besides, we have simulated our DSCPLD models using both full plant leaf images and segmented plant leaf images and conclude that, after using modified ACS, all models increase their accuracy and F1-score  ...  For this reason, initially, modified adaptive centroid-based segmentation (ACS) is used to trace the proper region of interest (ROI).  ...  A concrete representation of experiments on MobileNetV3 with width multipliers.Table 16. A concrete representation of experiments on MobileNetV3 with resolutions.  ... 
doi:10.3390/sym13030511 fatcat:2vljqd2wenerrlzzbr6kgcerhy

Discovering Multi-Hardware Mobile Models via Architecture Search [article]

Grace Chu, Okan Arikan, Gabriel Bender, Weijun Wang, Achille Brighton, Pieter-Jan Kindermans, Hanxiao Liu, Berkin Akin, Suyog Gupta, Andrew Howard
2021 arXiv   pre-print
single multi-hardware model yields similar or better results than SoTA performance on accelerators like GPU, DSP and EdgeTPU which was achieved by different models, while having similar performance with MobilenetV3  ...  Le, Hartwig Adam for helpful feedback and discussion; Cheng-Ming Chiang, Guan-Yu Chen, Koan-Sin Tan, Yu-Chieh Lin from MediaTek for useful guidance on Medi-aTek benchmarks; and QCT (Qualcomm CDMA Technologies  ...  Architecture search and training: We use ImageNet data [9] to search, train and evaluate. Input resolution is 224×224 and ResNet data preprocessing is used.  ... 
arXiv:2008.08178v2 fatcat:sg2qrauyjvesvpdjszt567o4aa

MUXConv: Information Multiplexing in Convolutional Neural Networks [article]

Zhichao Lu and Kalyanmoy Deb and Vishnu Naresh Boddeti
2020 arXiv   pre-print
On ImageNet, the resulting models, dubbed MUXNets, match the performance (75.3 of MobileNetV3 while being 1.6× more compact, and outperform other mobile models in all the three criteria.  ...  optimizing accuracy, compactness, and computational efficiency.  ...  This idea has also been particularly effective for image super-resolution [35] in the form of "subpixel" convolution.  ... 
arXiv:2003.13880v2 fatcat:shlwiywymve5zhfujzltoaotba

S CNet: Monocular Depth Completion for Autonomous Systems and 3D Reconstruction [article]

Lei Zhang, Weihai Chen, Chao Hu, Xingming Wu, Zhengguo Li
2019 arXiv   pre-print
In this paper, a lightweight yet efficient network (S\&CNet) is proposed to obtain a good trade-off between efficiency and accuracy for the dense depth completion.  ...  The CAM [26] pointed out the same issue and proposed a global variance pooling based SE module to improve their performance on the super-resolution.  ...  Most recently, MobileNetV3 [23] further improved the performance of efficient network by introducing the squeeze-and-excitation module.  ... 
arXiv:1907.06071v2 fatcat:xkjf776psjahdbylpfm26aeyu4
« Previous Showing results 1 — 15 out of 80 results