Training Data Selection for Depth Estimation Networks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The existing methods for collecting and storing images and point cloud data for depth estimation networks are costly and inefficient, often resulting in insufficient data collection, which hampers the improvement of depth estimation network performance, as they require predefined environmental conditions and do not account for varying conditions in real-time data collection.

Innovation Solution

A training data selection device that includes a depth estimation network, a vulnerability output device, and a training data acquisition support device, which applies depth estimation calculations to real-time input images, outputs depth distribution information, and determines depth estimation vulnerability to selectively store or transmit specific point cloud data when the vulnerability exceeds a threshold, thereby supplementing the training data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If all images and point cloud data collected in real time are stored, then sufficient training data is obtained, but huge storage costs are incurred

Engineering Contradiction:

Improvetraining data quantityVSAvoidstorage cost

Core Design Contradiction:

Quantity of substanceVSLoss of energy

Solution Approach 1:

The patent extracts only the essential information needed for training by using vulnerability assessment to identify and store only critical data points where the depth estimation network performs poorly, rather than storing all collected data. This extraction principle reduces storage requirements while maintaining training effectiveness.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter of data selection criteria from storing all data to storing only data exceeding a vulnerability threshold. By dynamically adjusting the selection parameter based on network performance assessment, the system reduces storage costs while ensuring sufficient training data quality.

Inventive Principle:

Principle #35Parameter changes

2Loss of energy

If predefined environmental conditions are used for data collection, then storage costs are reduced, but data necessary to improve network performance is insufficiently collected

Engineering Contradiction:

Improvestorage costVSAvoidtraining data quantity

Core Design Contradiction:

Loss of energyVSQuantity of substance

Solution Approach 1:

The patent implements a feedback mechanism where the depth estimation network continuously assesses its own performance on collected data, and this assessment feedback determines which data should be stored for training. The vulnerability assessment device provides feedback about data quality, enabling dynamic adjustment of storage decisions based on actual network performance rather than predefined conditions.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent performs preliminary vulnerability assessment of data before storage to identify data points that will most effectively improve network performance. This preliminary action ensures that storage resources are allocated to the most valuable data, preventing future performance deficiencies.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If depth estimation vulnerability is calculated for all input images, then accurate training data selection is achieved, but processing time is increased

Engineering Contradiction:

Improvedepth estimation accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by calculating vulnerability only for data that meets certain preliminary criteria or for a subset of data points, rather than processing all input images uniformly. This selective approach reduces processing time while maintaining sufficient accuracy for effective training data selection.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20240354975A1Training data selection device for selecting training data to improve performance of a depth estimation network and a training data selection method therefor

Publication Date: 2024.10.24 HYUNDAI MOTOR CO LTD

AI summary

A training data selection device for selecting training data and a training data selection method therefor are provided. The training data selection device includes a depth estimation network that applies depth estimation calculation to an input image obtained in real time to output depth distribution information corresponding to the input image. The device includes a vulnerability output device that outputs depth estimation vulnerability corresponding to the input image with reference to the depth distribution information. The device includes a training data acquisition support device that stores the input image and specific point cloud data corresponding to the input image as new training data in a certain storage space or transmits the input image and the specific point cloud data to another device, when it is determined that the depth estimation vulnerability is greater than or equal to a predetermined threshold.