Generalized Zero-Shot Vehicle Detection in Remote Sensing Imagery via Coarse-to-Fine Framework
Authors: Hong Chen, Yongtan Luo, Liujuan Cao, Baochang Zhang, Guodong Guo, Cheng Wang, Jonathan Li, Rongrong Ji
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results show that the proposed framework yields promising results on the imperative yet difficult task of zero-shot vehicle detection and recognition. In Sec. 4, we conduct quantitative experiments to verify the effectiveness of the proposed framework. |
| Researcher Affiliation | Collaboration | 1Fujian Key Laboratory of Sensing and Computing for Smart City, School of Information Science and Engineering, Xiamen University, China 2School of Automation Science and Electrical Engineering, Beihang University, China 3Institute of Deep Learning, Baidu Research 4National Engineering Laboratory for Deep Learning Technology and Application 5 Peng Cheng Laboratory, China |
| Pseudocode | Yes | Algorithm 1 Training procedure of the proposed model. |
| Open Source Code | No | The paper does not include an unambiguous statement or a direct link to the source code for the methodology described. |
| Open Datasets | Yes | Note that there is no dataset available in the task of generalized zero-shot vehicle detection of remote sensing images. We introduce a new dataset based on images from ISPRS WG III/4 2D Semantic Labeling Contest1. 1http://www2.isprs.org/commissions/comm3/wg4/2d-semlabel-potsdam.html |
| Dataset Splits | Yes | Data Split. We train the hierarchical Deep Lab v3 on the ISPRS 2D Semantic Labeling Contest dataset introduced above with the provided pixel-wise level annotations. This contest is to classify image pixels of remote sensing images into 6 categories, e.g., vehicle, tree and building. The zero-shot classification network is trained through the cropped vehicle patches. 12 categories of them are randomly selected for seen classes and the remaining 4 categories for unseen classes. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU models, CPU types) used to run the experiments. |
| Software Dependencies | No | The paper mentions models (VGG-16, Xception) and optimizers (SGD) but does not specify software dependencies with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x). |
| Experiment Setup | Yes | For all experiments, we use SGD optimizer with a momentum of 0.9 for training. Weight decay rate is fixed as 0.0005. α and β are tuned using five-fold cross-validation, while γ is fixed as 1. We tuned from 100 to 700 and fix the size of latent dictionary as 400 for better performance. |