Depth Estimation via Sub-Region Attention for Larger Receptive Fields
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional depth estimation models are limited by a small receptive field, leading to poor depth estimation results and inaccurate inference.
Innovation Solution
The method involves dividing an image into sub-regions, performing feature extraction on each sub-region, and using a pre-trained depth estimation model with a linear self-attention mechanism and multilayer perceptron to expand the receptive field and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a traditional convolutional neural network model is used for depth estimation, then the model structure remains simple, but the receptive field is limited and depth estimation accuracy deteriorates
Solution Approach 1:
The image is divided into multiple sub-regions, and each sub-region is processed independently to extract local features. This segmentation allows the model to capture detailed local information while the subsequent fusion of sub-region features expands the effective receptive field, resolving the contradiction between simple model structure and limited receptive field.
Solution Approach 2:
The patent introduces a new dimension by processing images in multiple sub-regions and then fusing the results. This dimensional approach (from processing the whole image at once to processing sub-regions and combining) effectively expands the receptive field without requiring a fundamentally more complex model architecture, thereby improving depth estimation accuracy while maintaining relative model simplicity.
2Measurement precision
If the model structure is kept simple, then ease of manufacture is maintained, but depth estimation accuracy and inference quality worsen
Solution Approach 1:
By segmenting the image into sub-regions and processing them separately, the model can achieve better depth estimation accuracy through localized feature extraction. The segmentation approach allows the use of relatively simple convolutional operations in each sub-region while the overall system achieves superior performance through the combination of multiple sub-region analyses.
Solution Approach 2:
The model applies convolutional operations multiple times across different sub-regions rather than using a single complex operation on the entire image. This partial action approach (processing parts separately and combining) achieves high accuracy without requiring an overly complex overall model structure, as each sub-region processing can use simpler operations.
Data Source
AI summary
The present application provides a depth estimation method, an electronic device and a storage medium, the method includes dividing an initial image into a plurality of sub-region images, and obtaining a feature vector corresponding to each sub-region image of the plurality of sub-region images by performing a feature extraction on each sub-region image. Once the feature vector corresponding to each sub-region image is input into a depth estimation model, and depth information corresponding to each feature vector is obtained using encoders of the depth estimation model, a depth image corresponding to the initial image is obtained using decoders of the depth estimation model based on the depth information corresponding to each feature vector. The present application can assist in a depth estimation and improve an accuracy of estimating a depth of an image.