ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2212.00770
8
0

On Utilizing Relationships for Transferable Few-Shot Fine-Grained Object Detection

1 December 2022
Ambar Pal
Arnau Ramisa
C. AmitKumarK
René Vidal
    ObjD
ArXivPDFHTML
Abstract

State-of-the-art object detectors are fast and accurate, but they require a large amount of well annotated training data to obtain good performance. However, obtaining a large amount of training annotations specific to a particular task, i.e., fine-grained annotations, is costly in practice. In contrast, obtaining common-sense relationships from text, e.g., "a table-lamp is a lamp that sits on top of a table", is much easier. Additionally, common-sense relationships like "on-top-of" are easy to annotate in a task-agnostic fashion. In this paper, we propose a probabilistic model that uses such relational knowledge to transform an off-the-shelf detector of coarse object categories (e.g., "table", "lamp") into a detector of fine-grained categories (e.g., "table-lamp"). We demonstrate that our method, RelDetect, achieves performance competitive to finetuning based state-of-the-art object detector baselines when an extremely low amount of fine-grained annotations is available (0.2%0.2\%0.2% of entire dataset). We also demonstrate that RelDetect is able to utilize the inherent transferability of relationship information to obtain a better performance (+5+5+5 mAP points) than the above baselines on an unseen dataset (zero-shot transfer). In summary, we demonstrate the power of using relationships for object detection on datasets where fine-grained object categories can be linked to coarse-grained categories via suitable relationships.

View on arXiv
Comments on this paper