Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Self-Prompting Analogical Reasoning for UAV Object Detection
Authors: Nianxin Li, Mao Ye, Lihua Zhou, Song Tang, Yan Gan, Zizhuo Liang, Xiatian Zhu
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments illustrate that SPAR outperforms traditional methods, offering a more robust and accurate solution for UAVOD. Experiment Experimental Settings Datasets. For the assement of our methods, we utilized two popular and challenging benchmark datasets in the field of aerial image detection: Vis Drone (Du et al. 2019) dataset and UAVDT (Du et al. 2018) dataset. Comparison with State-of-the-art Methods and Ablation Study sections with performance tables (Table 1, 2, 3) are also provided. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, University of Electronic Science and Technology of China, China 2Institute of Machine Intelligence (IMI), University of Shanghai for Science and Technology, China 3College of Computer Science, Chongqing University, China 4University of Sheffield 5University of Surrey EMAIL, EMAIL |
| Pseudocode | No | The paper describes its methodology using mathematical equations and descriptive text but does not include a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | No explicit statement or link is provided for the open-source code of the methodology described in this paper. |
| Open Datasets | Yes | For the assement of our methods, we utilized two popular and challenging benchmark datasets in the field of aerial image detection: Vis Drone (Du et al. 2019) dataset and UAVDT (Du et al. 2018) dataset. |
| Dataset Splits | Yes | Vis Drone dataset comprises 8599 images captured by drones, divided into 6471 for training, 548 for validation, and 1580 for testing, each with a resolution of approximately 2000 1500 pixels. UAVDT dataset is a comprehensive resource for drone-based tasks, including object detection, single-object tracking, and multiobject tracking. It comprises 24,143 training images and 16,592 testing images, each with an average resolution of 1024 540 pixels. |
| Hardware Specification | Yes | The backbone detector used in our study is YOLOv8, and all of our models use an NVIDIA RTX3090 GPU for training and testing. |
| Software Dependencies | No | The paper mentions 'YOLOv8' as the backbone detector but does not provide specific version numbers for YOLOv8 or any other software libraries, frameworks, or dependencies used. |
| Experiment Setup | Yes | In the training phase, we use part of pretrained model from YOLOv8, because SPAR and YOLOv8 share most part of backbone and some part of head, there are many weights can be transferred from YOLOv8x to our method, by using these weights we can save a lot of training time. The model on train set is trained for 100 epochs, and the first 2 epochs are used for warm-up. We use adam optimizer for training, and use 3e-4 as the initial learning rate. |