Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tiny Robotics Dataset and Benchmark for Continual Object Detection

Francesco Pasti, Riccardo De Monte, Davide Dalle Pezze, Gian Antonio Susto, Nicola Bellotto

TL;DR

The paper tackles continual object detection for tiny mobile robots under strict resource constraints by introducing the TiROD dataset and a continual learning benchmark. TiROD collects 6.7K frames across five environments with 13 classes, and the benchmark evaluates various CLOD methods on a lightweight detector (NanoDet), emphasizing domain and class incremental shifts. Empirical results show replay-based CL methods significantly outperform regularization-based approaches, with K-Means Replay delivering the best final performance, yet still below a cumulatively trained upper bound, highlighting the need for further advances in on-device continual learning. The work provides open data and code to spur development of robust, efficient incremental detectors for real-world tiny robotics applications.

Abstract

Detecting objects in mobile robotics is crucial for numerous applications, from autonomous navigation to inspection. However, robots often need to operate in different domains from those they were trained in, requiring them to adjust to these changes. Tiny mobile robots, subject to size, power, and computational constraints, encounter even more difficulties in running and adapting these algorithms. Such adaptability, though, is crucial for real-world deployment, where robots must operate effectively in dynamic and unpredictable settings. In this work, we introduce a novel benchmark to evaluate the continual learning capabilities of object detection systems in tiny robotic platforms. Our contributions include: (i) Tiny Robotics Object Detection~(TiROD), a comprehensive dataset collected using the onboard camera of a small mobile robot, designed to test object detectors across various domains and classes; (ii) a benchmark of different continual learning strategies on this dataset using NanoDet, a lightweight object detector. Our results highlight key challenges in developing robust and efficient continual learning strategies for object detectors in tiny robotics.

Tiny Robotics Dataset and Benchmark for Continual Object Detection

TL;DR

The paper tackles continual object detection for tiny mobile robots under strict resource constraints by introducing the TiROD dataset and a continual learning benchmark. TiROD collects 6.7K frames across five environments with 13 classes, and the benchmark evaluates various CLOD methods on a lightweight detector (NanoDet), emphasizing domain and class incremental shifts. Empirical results show replay-based CL methods significantly outperform regularization-based approaches, with K-Means Replay delivering the best final performance, yet still below a cumulatively trained upper bound, highlighting the need for further advances in on-device continual learning. The work provides open data and code to spur development of robust, efficient incremental detectors for real-world tiny robotics applications.

Abstract

Detecting objects in mobile robotics is crucial for numerous applications, from autonomous navigation to inspection. However, robots often need to operate in different domains from those they were trained in, requiring them to adjust to these changes. Tiny mobile robots, subject to size, power, and computational constraints, encounter even more difficulties in running and adapting these algorithms. Such adaptability, though, is crucial for real-world deployment, where robots must operate effectively in dynamic and unpredictable settings. In this work, we introduce a novel benchmark to evaluate the continual learning capabilities of object detection systems in tiny robotic platforms. Our contributions include: (i) Tiny Robotics Object Detection~(TiROD), a comprehensive dataset collected using the onboard camera of a small mobile robot, designed to test object detectors across various domains and classes; (ii) a benchmark of different continual learning strategies on this dataset using NanoDet, a lightweight object detector. Our results highlight key challenges in developing robust and efficient continual learning strategies for object detectors in tiny robotics.
Paper Structure (14 sections, 1 equation, 3 figures, 2 tables)

This paper contains 14 sections, 1 equation, 3 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Tiny Robotics
  4. Continual Learning
  5. Related Datasets
  6. TiROD DATASET
  7. Tiny Mobile Robot
  8. Dataset Details
  9. Continual Learning Benchmark
  10. BENCHMARK
  11. Implementation Details and CLOD Methods
  12. Evaluation Metrics
  13. Results
  14. CONCLUSIONS

Figures (3)

  • Figure 1: TiROD: Tiny Robotics Object Detection dataset, with a tiny mobile robot exploring multiple environments. It must learn to detect objects in each new domain without forgetting the previously acquired knowledge.
  • Figure 2: TiROD dataset samples. Each column corresponds to one of the 10 Continual Learning tasks. Every two tasks there is a domain change while for each domain there are two illumination conditions, "High" and "Low".
  • Figure 3: TiROD dataset details. Vertical dotted lines mark the separation between learning tasks. The top two rows represent the illumination level. The categories plots are histograms representing the number of instances for each class every 100 frames. Histogram scales per category are shown on the right.