Image Dataset Cleaning with IoU Filtering for Redundant Frame Removal

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for managing large volumes of image data in automation technology, particularly in automatic unpacking machines and autonomous vehicles, struggle with semantic redundancy, leading to inefficient data processing and loss of relevant information.

Innovation Solution

A method utilizing an intersection-over-union (IoU) filter to identify and remove redundant images by comparing keyframes with a predetermined threshold, focusing on foreground changes and maintaining dataset quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If semantic redundancy filtering is applied to reduce data volume, then processing efficiency improves, but risk of losing relevant information increases

Engineering Contradiction:
Improvedata processing efficiencyVSAvoidloss of relevant information
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent replaces traditional mechanical redundancy filtering methods with a deep learning-based semantic analysis system. The neural network model analyzes the semantic content of images to identify and remove redundant data, substituting mechanical filtering with intelligent semantic understanding to maintain information quality while improving processing efficiency

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the parameter of redundancy detection from simple visual similarity to semantic content similarity. By using deep learning models to analyze semantic parameters rather than basic image parameters, the system can more accurately distinguish between redundant and valuable data, reducing information loss while maintaining high processing efficiency

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If deep learning models are trained on large volumes of image data, then model accuracy improves, but training time and computational resources increase

Engineering Contradiction:
Improvemodel classification accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts and removes redundant images from the training dataset using semantic redundancy filtering. By taking out duplicate and highly similar images that provide minimal additional learning value, the system reduces the total training data volume while preserving the most informative samples, thereby decreasing training time without significantly compromising model accuracy

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent performs preliminary redundancy filtering on the image dataset before training the deep learning model. This preliminary action of removing redundant data beforehand reduces the computational burden during training, allowing the model to learn from high-quality, non-redundant samples more efficiently, thus reducing training time while maintaining accuracy

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4654154A1Method and device for cleaning a set of image data
Publication Date: 2025.11.26 ROBERT BOSCH GMBH

AI summary

A method and apparatus (100) for cleaning an image data set used for training and/or validating and/or testing a machine learning model is proposed, the method comprising the steps of: - providing (S1) the image data set, comprising a plurality of images; - comparing (S2) a, in particular predetermined, comparison image from the plurality of images with at least a subset of the remaining images of the plurality of images by applying an intersection-over-union filter; - based on the comparison, identifying (S3) at least one image redundant with respect to the comparison image in at least the subset of the remaining images of the plurality of images; and - cleaning (S4) the image data set by removing the at least one redundant image from the plurality of images.