Instance segmentation algorithm based on peak response enhancement and computing equipment

Abstract

The invention discloses an instance segmentation algorithm based on peak response enhancement and computing equipment, and the algorithm comprises the steps: firstly obtaining a data set which comprises a training image and a test image, and only the training image comprises the class information of an object on the image; designing an implementation peak response generator to acquire a corresponding incomplete peak response graph based on the input image and the implementation peak response generator; designing a peak response enhancement network, and based on the incomplete peak response graph and the peak response enhancement network to acquire an enhanced version peak response graph with more comprehensive coverage; performing physical property-like sampling on the input image to obtain a physical property-like sampling set; designing a matching strategy; and finally, according to a matching strategy, matching the enhanced version peak response graph corresponding to the test imagewith the physical property sampling set to obtain a final instance segmentation result of the test image. According to the invention, the weak supervision instance segmentation effect is realized.

Description

technical field [0001] The invention relates to the technical field of weakly supervised instance segmentation, in particular to an instance segmentation algorithm and computing equipment based on peak response enhancement. Background technique [0002] Advances in image recognition technology are driving developments in fields ranging from autonomous driving to medical diagnostics. Many researchers are racing to develop deep learning algorithms for object recognition that enable high-precision real-time object detection and classification, many of which rely on image segmentation techniques powered by convolutional neural networks, which make up the computer Fundamentals of Deep Learning for Vision. Image segmentation is actually segmenting the boundaries of objects in the input image at the pixel level, which helps to achieve object detection tasks in real-world scenes and helps to distinguish multiple similar objects in the same image. Image segmentation can be subdivid...

AI Technical Summary

Problems solved by technology

Although the peak response maps highlight the most obvious parts of objects, through which we can roughly locate instances of each class, such incomplete peak response maps cannot provide reliable information for segmentation, which is Because the incomplete peak response plot does not know which region can be considered a complete instance

It can be seen that the existing methods cannot achieve accurate instance segmentation, so it is necessary to develop new methods that can achieve accurate instance segmentation

Images