Self‐supervised monocular depth estimation via asymmetric convolution block

Without the dependence of depth ground truth,self‐supervised learning is a promising alternative to train monocular depth estimation.It builds its own supervision signal with the help of other tools,such as view synthesis and pose networks.However,more training parameters and time consumption may be involved.This paper proposes a monocular depth prediction framework that can jointly learn the depth value and pose transformation between images in an end‐to‐end manner.The depth network creatively employs an asymmetric convolution block instead of every square kernel layer to strengthen the learning ability of extracting image features when training.During infer-ence time,the asymmetric kernels are fused and converted to the original network to predict more accurate image depth,thus bringing no extra computations anymore.The network is trained and tested on the KITTI monocular dataset.The evaluated results demonstrate that the depth model outperforms some State of the Arts(SOTA)ap-proaches and can reduce the inference time of depth prediction.Additionally,the pro-posed model performs great adaptability on the Make3D dataset.

A Novel Fish Counting Method Based on Multiscale and Multicolumn Convolution Group Network

An accurate grasp of the number of fish in the breeding pond or fixed waters can provide an important basis for bait placement and reasonable fishing,and these data can also provide the necessary data support for accurate breeding.Due to the high density of fish in the real underwater environment,the strong occlusion and the large amount of adhesion,it is difficult to count fish,and the accuracy is low.Considering the above issues,we present a new approach to a fish counting method based on a multiscale multicolumn convolution group network.To enhance the counting accuracy and reduce the complexity of the network,this method uses an asymmetric convolution kernel to change the traditional convolution kernel,which increases our network depth and appreciably reduces the size of the network.In the backbone network,a convolutional group is used to replace a single convolutional layer to enhance the learning capacity of the network.The back of the net introduces the spatial structure of the pyramid and the multicolumn dilated convolution,which preserves the different scaling properties of fish data and improves the capabilities of the fish counting algorithm.To check the performance of the algorithm,this work collects and labels the DLOU3 fish dataset suitable for counting fish and conducts simulation experiments on the DLOU3 fish dataset using our algorithm.The experiments are comparedwith other popular fish counting algorithms in terms of the mean absolute error(MAE)and mean square error(MSE).The MAE and MSE of the final experimental results of our method are 5.36 and 6.56 and 23.67 and 32.52 in the two test sets,respectively,and the best performance among the five groups of algorithms is obtained.