Search

Traditional model compression techniques are dependent on handcrafted features and require domain experts, with a tradeoff between model size, speed, and accuracy. This study proposes a new approach toward resolving model compression problems. Our approach combines reinforcement-learning-based automated pruning and knowledge distillation to improve the pruning of unimportant network layers and the efficiency of the compression process. We introduce a new state quantity that controls the size of the reward and an attention mechanism that reinforces useful features and attenuates useless features to enhance the effects of other features. The experimental results show that the proposed model is superior to other advanced pruning methods in terms of the computation time and accuracy on CIFAR-100 and ImageNet dataset, where the accuracy is approximately 3% higher than that of similar methods with shorter computation times. [ABSTRACT FROM AUTHOR]

Depth images obtained from lightweight, real-time depth estimation models and consumer-oriented sensors typically have low-resolution issues. Traditional interpolation methods for depth image up-sampling result in a significant information loss, especially in edges with discontinuous depth variations (depth discontinuities). To address this issue, this paper proposes a semi-coupled deformable convolution network (SCD-Net) based on the idea of guided depth map super-resolution (GDSR). The method employs a semi-coupled feature extraction scheme to learn unique and similar features between RGB images and depth images. We utilize a Coordinate Attention (CA) to suppress redundant information in RGB features. Finally, a deformable convolutional module is employed to restore the original resolution of the depth image. The model is tested on NYUv2, Middlebury, Lu, and a Real-Sense real-world dataset created using an Intel Real-sense D455 structured-light camera. The super-resolution accuracy of SCD-Net at multiple scales is much higher than that of traditional methods and superior to recent state-of-the-art (SOTA) models, which demonstrates the effectiveness and flexibility of our model on GDSR tasks. In particular, our method further solves the problem of an RGB texture being over-transferred in GDSR tasks. [ABSTRACT FROM AUTHOR]

To solve the problem of complex network models with a large number of redundant parameters, a pruning algorithm combined with an attention mechanism is proposed. Firstly, the basic training is performed once, and the network model is then re-trained with the attention mechanism for the baseline. The obtained model is pruned based on channel correlation, and finally a simplified model is obtained via continuous cyclic iteration while the accuracy rate is kept as close as possible to that of the baseline model. The algorithm was experimentally validated on ResNet based on different datasets, and the results showed that the algorithm provided strong adaptability to different datasets and different network structures. For the CIFAR-100 dataset, ResNet50 was pruned to reduce the amount of model parameters by 80.3% and the amount of computation by 69.4%, while maintaining accuracy. For the ImageNet dataset, the ResNet50 parameter volume was compressed by 2.49 times and the computational volume was compressed by 3.01 times. The ResNet101 parameter volume was reduced by 61.2%, and the computational volume was reduced by 68.5%. Compared with the traditional fixed threshold, the model achieves better results in terms of detection accuracy, compression effect, and inference speed. [ABSTRACT FROM AUTHOR]

In the single pruning algorithm, channel pruning or filter pruning is used to compress the deep convolution neural network, and there are still many redundant parameters in the compressed model. Directly pruning the filter will largely cause the loss of key information and affect the accuracy of model classification. To solve these problems, a parallel pruning algorithm combined with image enhancement is proposed. Firstly, in order to improve the generalization ability of the model, a data enhancement method of random erasure is introduced. Secondly, according to the trained batch normalization layer scaling factor, the channels with small contribution are cut off, the model is initially thinned, and then the filters are pruned. By calculating the geometric median of the filters, redundant filters similar to them are found and pruned, and their similarity is measured by calculating the distance between filters. Pruning was done using VGG19 and DenseNet40 on cifar10 and cifar100 data sets. The experimental results show that this algorithm can improve the accuracy of the model, and at the same time, it can compress the calculation and parameters of the model to a certain extent. Finally, this method is applied in practice, and combined with transfer learning, traffic objects are classified and detected on the mobile phone. [ABSTRACT FROM AUTHOR]

In order to avoid the direct depth reconstruction of the original image pair and improve the accuracy of the results, we proposed a coarse-to-fine stereo matching network combining multi-level residual optimization and depth map super-resolution (ASR-Net). First, we used the u-net feature extractor to obtain the multi-scale feature pair. Second, we reconstructed global disparity in the lowest resolution. Then, we regressed the residual disparity using the higher-resolution feature pair. Finally, the lowest-resolution depth map was refined by using the disparity residual. In addition, we introduced deformable convolution and group-wise cost volume into the network to achieve adaptive cost aggregation. Further, the network uses ABPN instead of the traditional interpolation method. The network was evaluated on three datasets: scene flow, kitti2015, and kitti2012 and the experimental results showed that the speed and accuracy of our method were excellent. On the kitti2015 dataset, the three-pixel error converged to 2.86%, and the speed was about six times and two times that of GC-net and GWC-net. [ABSTRACT FROM AUTHOR]

To further reduce the size of the neural network model and enable the network to be deployed on mobile devices, a novel fusion pruning algorithm based on information entropy stratification is proposed in this paper. Firstly, the method finds similar filters and removes redundant parts by Affinity Propagation Clustering, then secondly further prunes the channels by using information entropy stratification and batch normalization (BN) layer scaling factor, and finally restores the accuracy training by fine-tuning to achieve a reduced network model size without losing network accuracy. Experiments are conducted on the vgg16 and Resnet56 network using the cifar10 dataset. On vgg16, the results show that, compared with the original model, the parametric amount of the algorithm proposed in this paper is reduced by 90.69% and the computation is reduced to 24.46% of the original one. In ResNet56, we achieve a 63.82%-FLOPs reduction by removing 63.53% parameters. The memory occupation and computation speed of the new model are better than the baseline model while maintaining a high network accuracy. Compared with similar algorithms, the algorithm has obvious advantages in the dimensions of computational speed and model size. The pruned model is also deployed to the Internet of Things (IoT) as a target detection system. In addition, experiments show that the proposed model is able to detect targets accurately with low reasoning time and memory. It takes only 252.84 ms on embedded devices, thus matching the limited resources of IoT. [ABSTRACT FROM AUTHOR]

In order to solve the problem of large model computing power consumption, this paper proposes a novel model compression algorithm. Firstly, this paper proposes an interpretable weight allocation method for the loss between a student network (a network model with poor performance), a teacher network (a network model with better performance) and real label. Then, different from the previous simple pruning and fine-tuning, this paper performs knowledge distillation on the pruned model, and quantifies the residual weights of the distilled model. The above operations can further reduce the model size and calculation cost while maintaining the model accuracy. The experimental results show that the weight allocation method proposed in this paper can allocate a relatively appropriate weight to the teacher network and real tags. On the cifar-10 dataset, the pruning method combining knowledge distillation and quantization can reduce the memory size of resnet32 network model from 3726 KB to 1842 KB, and the accuracy can be kept at 93.28%, higher than the original model. Compared with similar pruning algorithms, the model accuracy and operation speed are greatly improved. [ABSTRACT FROM AUTHOR]