Grid Anchor based Image Cropping Database (GAICD)
收藏资源简介:
GAICD是由香港理工大学计算机系构建的一个新型图像裁剪数据库,包含1,236张图像和106,860个标注的候选裁剪。该数据集通过考虑图像裁剪的特殊属性和需求,如局部冗余、内容保护和宽高比,采用网格锚点方法大幅减少了候选裁剪的数量,从数百万减少到不超过90个。数据集中的每个裁剪都由7位经验丰富的人类受试者进行标注,提供了丰富的监督信息,并定义了三种新的评估指标,以更可靠地评估裁剪模型的性能。GAICD旨在为研究人员提供一个更合理的裁剪基准,以开发和评估他们的模型,特别适用于解决图像美学和构图优化的问题。
GAICD is a novel image cropping database constructed by the Department of Computer Science, The Hong Kong Polytechnic University. It contains 1,236 images and 106,860 annotated candidate croppings. By considering the special attributes and requirements of image cropping such as local redundancy, content preservation and aspect ratio, the grid anchor method is adopted to drastically reduce the number of candidate croppings from millions to no more than 90. Each cropping in the dataset is annotated by 7 experienced human subjects, providing rich supervisory information. Three novel evaluation metrics are defined to more reliably evaluate the performance of cropping models. GAICD aims to provide researchers with a more reasonable cropping benchmark for developing and evaluating their models, and is particularly suitable for addressing image aesthetics and composition optimization problems.
Grid-Anchor-based-Image-Cropping-Pytorch 数据集概述
环境要求
- Python 2.7
- PyTorch 0.4.1
- NumPy
- OpenCV (cv2)
- SciPy
使用方法
-
数据集下载:
-
预训练模型下载:
-
运行代码:
- 训练新模型:运行
TrainModel.py - 测试预训练模型:运行
demo_eval.py
- 训练新模型:运行
-
修改裁剪比例:
- 在
croppingDataset.py文件中修改generate_bboxes函数(第115行)
- 在
其他实现
引用
@inproceedings{zhang2019deep, title={Reliable and Efficient Image Cropping: A Grid Anchor based Approach}, author={Zeng, Hui, Li, Lida, Cao, Zisheng and Zhang, Lei}, booktitle={IEEE Conference on Computer Vision and Pattern Recognition}, year={2019} } @article{zeng2020cropping, title={Grid Anchor based Image Cropping: A New Benchmark and An Efficient Model}, author={Zeng, Hui and Li, Lida and Cao, Zisheng and Zhang, Lei}, journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, volume={}, number={}, pages={}, year={2020}, publisher={IEEE} }

- 1Grid Anchor based Image Cropping: A New Benchmark and An Efficient Model香港理工大学计算机系 · 2019年



