Abstract:Weakly supervised object detection (WSOD) is a challenging task, in which image-level labels (e.g., categories of the instances in the whole image) are used to train an object detector. Many existing methods follow the standard multiple instance learning (MIL) paradigm and have achieved promising performance. However, the lack of deterministic information leads to part domination and missing instances. To address these issues, this paper focuses on identifying and fully exploiting the deterministic information in WSOD. We discover that negative instances (i.e. absolutely wrong instances), ignored in most of the previous studies, normally contain valuable deterministic information. Based on this observation, we here propose a negative deterministic information (NDI) based method for improving WSOD, namely NDI-WSOD. Specifically, our method consists of two stages: NDI collecting and exploiting. In the collecting stage, we design several processes to identify and distill the NDI from negative instances online. In the exploiting stage, we utilize the extracted NDI to construct a novel negative contrastive learning mechanism and a negative guided instance selection strategy for dealing with the issues of part domination and missing instances, respectively. Experimental results on several public benchmarks including VOC 2007, VOC 2012 and MS COCO show that our method achieves satisfactory performance.

What problem does this paper attempt to address?

This paper attempts to solve two main problems in Weakly Supervised Object Detection (WSOD): **discriminative part domination** and **missing instances**. ### 1. Discriminative Part Domination In weakly - supervised object detection, due to the lack of accurate instance - level annotation information, existing methods can usually only construct image classification tasks to provide supervision information. This causes the model to often focus too much on the discriminative parts of the target (such as the head of a dog or the beak of a bird), while ignoring other parts, making it difficult to determine whether a bounding box contains a complete target object, especially when dealing with non - rigid objects (such as cats and dogs). ### 2. Missing Instances Existing methods usually generate pseudo - labels by selecting the candidate box with the highest confidence for training the instance refinement branch, but this method ignores other potential target objects, leading to the problem of missed detection. For example, when selecting the proposal with the highest confidence as a positive sample, other possible targets may be ignored, thus affecting the comprehensiveness and accuracy of detection. ### Solution To solve the above problems, the author proposes a method based on Negative Deterministic Information (NDI), namely NDI - WSOD. Specifically: - **NDI Collection Phase**: By identifying and extracting negative deterministic information in negative samples (i.e., absolutely wrong instances), these negative samples usually contain valuable deterministic information. - **NDI Utilization Phase**: - **Negative Contrastive Learning (NCL)**: Use the collected NDI to construct a negative contrastive learning mechanism, reduce the over - fitting of discriminative parts, and help the model jump out of the local optimal solution. - **Negative Guided Instance Selection (NGIS)**: Use NDI as a template to screen out more foreground instances, avoid noise interference, and ensure recall rate and precision. Through these two phases, NDI - WSOD can effectively alleviate the problems of discriminative part domination and missing instances, thereby improving the performance of weakly - supervised object detection. ### Experimental Results The experimental results show that NDI - WSOD has achieved significant performance improvements on multiple public datasets (such as VOC 2007, VOC 2012, and MS COCO), verifying the effectiveness of this method. In summary, this paper proposes a novel weakly - supervised object detection method by introducing negative deterministic information, which effectively solves the problems of discriminative part domination and missing instances in existing methods.

Absolute Wrong Makes Better: Boosting Weakly Supervised Object Detection via Negative Deterministic Information

Negative Deterministic Information-Based Multiple Instance Learning for Weakly Supervised Object Detection and Segmentation

Spatial Likelihood Voting with Self-Knowledge Distillation for Weakly Supervised Object Detection.

Refining and reweighting pseudo labels for weakly supervised object detection

Negative Prototypes Guided Contrastive Learning for WSOD

USD: Unknown Sensitive Detector Empowered by Decoupled Objectness and Segment Anything Model

Learning Hybrid Negative Probability Model for Weakly-Supervised Whole Slide Image Recognition.

MOL: Towards Accurate Weakly Supervised Remote Sensing Object Detection Via Multi-view Noisy Learning

Instance-Level Contrastive Learning for Weakly Supervised Object Detection

Towards Precise Weakly Supervised Object Detection via Interactive Contrastive Learning of Context Information

A Dual-Network Progressive Approach to Weakly Supervised Object Detection.

Attention Erasing and Instance Sampling for Weakly Supervised Object Detection

Discovery-and-Selection: Towards Optimal Multiple Instance Learning for Weakly Supervised Object Detection

Optimizing Region Selection for Weakly Supervised Object Detection

Dynamic sample weighting for weakly supervised object detection

Few-shot Weakly-Supervised Object Detection via Directional Statistics

Weakly Supervised Object Detection for Remote Sensing Images via Progressive Image-Level and Instance-Level Feature Refinement

Recurrent Self-Optimizing Proposals for Weakly Supervised Object Detection

Salvage of Supervision in Weakly Supervised Detection

High-Quality Proposals for Weakly Supervised Object Detection.

From Discriminant to Complete: Reinforcement Searching-Agent Learning for Weakly Supervised Object Detection