UN-DETR: Promoting Objectness Learning via Joint Supervision for Unknown Object Detection

Authors: HaoMiao Liu, Hao Xu, Chuhuai Yue, Bo Ma

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our UN-DETR is comprehensively evaluated on multiple UOD and known detection benchmarks, demonstrating its effectiveness and achieving state-of-the-art performance. Experiment Following the UOD Benchmark, we utilize COCO-OOD, COCO-Mixed (Liang et al. 2023), and VOC (Everingham et al. 2010) as test sets and employ m AP, U-AP, U-F1, U-PRE, and U-REC as evaluation metrics, as detailed in the Appendix. Tables 1 and 2 present the results of our method UN-DETR, alongside 8 classic or recent stateof-the-art methods, on the UOD Benchmark. To examine the contribution of each component in our method, we conduct adequate ablation experiments as presented in Table 3.
Researcher Affiliation Academia Haomiao Liu*, Hao Xu*, Chuhuai Yue*, Bo Ma , Beijing Institute of Technology EMAIL
Pseudocode No The paper describes methods and processes in narrative text and figures, but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/ndwxhmzz/UN-DETR
Open Datasets Yes Experiment Following the UOD Benchmark, we utilize COCO-OOD, COCO-Mixed (Liang et al. 2023), and VOC (Everingham et al. 2010) as test sets and employ m AP, U-AP, U-F1, U-PRE, and U-REC as evaluation metrics
Dataset Splits No The paper uses established benchmark datasets and refers to "test sets" (COCO-OOD, COCO-Mixed, VOC) and "training set" (VOC training set for pretraining), implying standard splits. However, it does not provide explicit percentages, sample counts, or a detailed methodology for splitting beyond stating which datasets are used for training and testing.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory amounts used for running the experiments.
Software Dependencies No The paper does not provide specific software dependency details, such as library names with version numbers (e.g., PyTorch 1.9, CUDA 11.1). It mentions using ResNet50 as a backbone, which is a model architecture, not a software dependency with a specific version.
Experiment Setup Yes The weight parameters α and β are empirically set to 0.6 and 0.4, respectively. In Eq. 4, C is set to 0.5 and τ is set to 0.6. ... λ1, λ2, and λ3 are the weights of the loss, set to 3, 2, and 5, respectively.