Slim-CNN: A Light-Weight CNN for Face Attribute Prediction

Sharma, Ankit Kumar; Foroosh, Hassan

doi:10.1109/fg47880.2020.00085

Cited by 42 publications

(15 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we compare the proposed SSPL method with ten state-of-the-art methods, including five supervised FAR methods [22,27,20,2,14], three self-supervised learning methods [3,24,10], and two semi-supervised learning methods [28,23], on the CelebA and LFWA 2.…”

Section: Comparison With State-of-the-art Methodsmentioning

confidence: 99%

“…With the increasing availability of large-scale data, deep learning-based methods have become dominant in the field of FAR. Sharma and Foroosh [27] leverage deep separable convolutions and pointwise convolution to design a lightweight CNN for FAR, which significantly reduces the model parameters and improves the computational efficiency. Mao et al [22] perform FAR based on a Deep Multi-task and Multi-label Convolutional Neural Network (DMM-CNN).…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Learning Spatial-Semantic Relationship for Facial Attribute Recognition with Limited Labeled Data

Ying

Yan

Chen

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Section: Comparison With State-of-the-art Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Learning Spatial-Semantic Relationship for Facial Attribute Recognition with Limited Labeled Data

Ying

Yan

Chen

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

“…To diagnose the PV module fault type by images, this study employed the CDEM image (containing information of the fault state) based on the features generated by the Lorenz master and slave chaotic systems [21] in CSDM for fault recognition by the CNN. The CNN is a part that is defined in a deep learning network in modern times, e.g., facial feature recognition [22], biometric recognition embedded in an FPGA system [23] and high-tension cable fault diagnosis [24]. It is extensively used in signal processing and image classification.…”

Section: Convolutional Neural Networkmentioning

confidence: 99%

Photovoltaic Module Fault Detection Based on a Convolutional Neural Network

Wang²,

Wei³

et al. 2021

Processes

View full text Add to dashboard Cite

With the rapid development of solar energy, the photovoltaic (PV) module fault detection plays an important role in knowing how to enhance the reliability of the solar photovoltaic system and knowing the fault type when a system problem occurs. Therefore, this paper proposed the hybrid algorithm of chaos synchronization detection method (CSDM) with convolutional neural network (CNN) for studying PV module fault detection. Four common PV module states were discussed, including the normal PV module, module breakage, module contact defectiveness and module bypass diode failure. First of all, the defects in 16 pieces of 20W monocrystalline silicon PV modules were preprocessed, and there were four pieces of each fault state. When the signal generator delivered high frequency voltage to the PV module, the original signal was measured and captured by the NI PXI-5105 high-speed data acquisition system (DAS) and was calculated by CSDM, to establish the chaos dynamic error map as the image feature of fault diagnosis. Finally, the CNN was employed for diagnosing the fault state of the PV module. The findings show that after entering 400 random fault data (100 data for each fault) into the proposed method for recognition, the recognition accuracy rate of the proposed method was as high as 99.5%, which is better than the traditional ENN algorithm that had a recognition rate of 86.75%. In addition, the advantage of the proposed algorithm is that the mass original measured data can be reduced by CSDM, the subtle changes in the output signals are captured effectively and displayed in images, and the PV module fault state is accurately recognized by CNN.

show abstract

“…For each setting, we measure the efficacy of the edits on a sample of 10, 000 generated images, and we also quantify the undesired changes made by each method. For the smiling edit, we measure efficacy by counting images classified as smiling by an attribute classifier [69], and we also quantify changes made in the images outside the mouth region by masking lips using a face segmentation model [87] and using LPIPS [84] to quantify changes. For the dome edits, we Table 3: Visible watermark text produced by StyleGANv2 church model in n = 1000 images, without modification, with sets of units zeroed (using the method of GAN Dissection), and using our method to apply a rank-one update.…”

Section: Putting Objects Into a New Contextmentioning

confidence: 99%

“…To quantify the efficacy of the change, we also use pretrained networks. To detect whether a face image is similing, we use a Slim-CNN [69] facial attribute classifier. To determine if domes have successfully been edited to other types of objects, we again use the Unified Perceptual Parsing network, and we count pixels that have changed from being classified as domes to buildings or trees.…”

Section: Metricsmentioning

confidence: 99%

Rewriting a Deep Generative Model

Bau¹,

Liu²,

Wang³

et al. 2020

Preprint

View full text Add to dashboard Cite

A deep generative model such as a GAN learns to model a rich set of semantic and physical rules about the target distribution, but up to now, it has been obscure how such rules are encoded in the network, or how a rule could be changed. In this paper, we introduce a new problem setting: manipulation of specific rules encoded by a deep generative model. To address the problem, we propose a formulation in which the desired rule is changed by manipulating a layer of a deep network as a linear associative memory. We derive an algorithm for modifying one entry of the associative memory, and we demonstrate that several interesting structural rules can be located and modified within the layers of state-ofthe-art generative models. We present a user interface to enable users to interactively change the rules of a generative model to achieve desired effects, and we show several proof-of-concept applications. Finally, results on multiple datasets demonstrate the advantage of our method against standard fine-tuning methods and edit transfer algorithms.

show abstract

Slim-CNN: A Light-Weight CNN for Face Attribute Prediction

Cited by 42 publications

References 17 publications

Learning Spatial-Semantic Relationship for Facial Attribute Recognition with Limited Labeled Data

Learning Spatial-Semantic Relationship for Facial Attribute Recognition with Limited Labeled Data

Photovoltaic Module Fault Detection Based on a Convolutional Neural Network

Rewriting a Deep Generative Model

Contact Info

Product

Resources

About