Fast-Zero-Shot-Image-Tagging-master
所属分类：图形图像处理
开发工具：Python
文件大小：2623KB
下载次数：12
上传日期：2018-01-08 15:11:19
上 传 者：
899
说明： LINUX环境，对图像中物体打标签，深度学习。
(In LINUX,tagging labels of the object in the image, deep learning.)
文件列表:
Dataset (0, 2017-11-11)
Dataset\downloader.py (571, 2017-11-11)
Dataset\labels_test.zip (1081036, 2017-11-11)
Dataset\labels_train.zip (1607653, 2017-11-11)
fast0tag.py (853, 2017-11-11)
ranking.py (5465, 2017-11-11)
# Fast Zero-Shot Image Tagging (in progress)
Implementing algorithm from paper by Zhang , et al
Paper: http://crcv.ucf.edu/papers/cvpr2016/Zhang_CVPR2016.pdf
NUS-WIDE Dataset: http://lms.comp.nus.edu.sg/research/NUS-WIDE.htm
/Dataset contains compressed files for constructing dataframes for the NUS-WIDE dataset. For the raw data, download directly from the link above.
point /Dataset/downloader.py to the appropriate file containing img urls to download images.
**There are a large number of broken links in NUS-WIDE-urls.text if it is crucial to have access to all 260k images, request access from the authors of the dataset at the NUS-WIDE Dataset website.**
## Naming conventions used by authors of NUS-WIDE dataset
- _Concepts_ are the image classes. There is some overlap in the classes corresponding to fine-grained image classification (for example there exists __animal__ and __bird__ classes).
- _Groundtruth_ are solutions to both training and test sets for assigning concept to image.
- _Tags_ are more general categorical assignments with a high degree of overlap between tags.
近期下载者：
相关文件：
收藏者：