yty3805595

Pinned Repositories

A-to-Z-Resources-for-Students
Curated list of resources for college students
0 1 00
adversarial-frcnn-master
基于GAN的目标检测
Language:Python0 1 00
caffe
Caffe: a fast open framework for deep learning.
Language:C++0 1 00
caffe-windows
Configure Caffe in one hour for Windows users.
Language:C++00
chatbot-base-on-Knowledge-Graph
使用深度学习方法解析问题知识图谱存储查询知识点基于医疗垂直领域的对话系统
Language:Python00
ChineseCodingInterviewAppendix
The source code for the appendix part of the Chinese version of the book Coding Interviews
Language:C++0 1 00
classify
animal classify
Language:Python0 2 00
cv-papers
计算机视觉相关论文整理、记录、分享; 包括图像分类、目标检测、视觉跟踪/目标跟踪、人脸识别/人脸验证、OCR/场景文本检测及识别等领域。欢迎加星,欢迎指正错误,同时也期待能够共同参与！！! 持续更新中... ...
00
CV_Homework
Language:Jupyter Notebook0 1 00
typhoon_prediction
# typhoon Analysis satellite images of typhoons by deep-learning (CNN), based on PyTorch. This CNN learns to map the satellite images of typhoons to their max wind speed from. The labeled train set is obtained from agora/JMA. ## Requirements * BeautifulSoup * PIL * Pytorch ## Usage 1. Run `download.py`, download the satellite images of typhoons to folder `tys_raw`. 2. Run `create_samples.py`, convert raw data into the legal samples for our CNN, create two new forlder `train_set` and `test_set`. 3. Train CNN using `train_net.py`, the trained CNN will be saved as a disk file `net_relu.pt`. 4. Run `test_net.py`, analysis the test set. After 10 epoches training the CNN regressor reached mean loss about 8 (knots) in train set and about 10 (knots) in test set. ![](https://raw.githubusercontent.com/melissa135/deep_typhoon/master/loss_sequence.png) Here is what this CNN thinks of the top 20 typhoons sorted by max wind. ``` 1 ('197920', 130.27679443359375) 2 ('200914', 127.7662582397461) 3 ('199019', 122.92172241210938) 4 ('200918', 122.84004211425781) 5 ('201614', 122.66597747802734) 6 ('201601', 122.03250885009766) 7 ('201513', 121.75947570800781) 8 ('200922', 121.35771942138672) 9 ('201013', 120.0194091796875) 10 ('201330', 118.92587280273438) 11 ('201419', 117.6025390625) 12 ('198305', 117.10270690917969) 13 ('201422', 116.77259063720703) 14 ('198522', 116.46116638183594) 15 ('201327', 116.42304992675781) 16 ('201216', 116.36921691894531) 17 ('198221', 116.18096923828125) 18 ('199230', 115.96656799316406) 19 ('198210', 115.96611022949219) 20 ('201328', 115.57132720947266) ``` ## Tips * Memory should be at least 1.5G . * This project is written without `cuda()`, while you can use `cuda()` to transfer the CNN onto GPU and speedup the training. * The images and labels are crawled from agora.ax.nii.ac.jp/digital-typhoon , and the labels are refered to JMA(Japan Meteorological agency).
Language:Python16 2 04

yty3805595's Repositories

yty3805595/adversarial-frcnn-master
基于GAN的目标检测
Language:Python0 1 00
yty3805595/caffe-windows
Configure Caffe in one hour for Windows users.
Language:C++00
yty3805595/ChineseCodingInterviewAppendix
The source code for the appendix part of the Chinese version of the book Coding Interviews
Language:C++0 1 00
yty3805595/Deedy-Resume-for-Chinese
适合应届毕业生的 LaTeX 简历模板
Language:TeX
yty3805595/deep_typhoon
Analysis satellite images of typhoons in deep-learning (CNN).
Language:Python
yty3805595/hourglass-facekeypoints-detection
face keypoints deteciton based on stackedhourglass
Language:Python
yty3805595/PaintsChainer
line drawing colorization using chainer
yty3805595/The-Little-Heat-Design
散热设计学习小书
yty3805595/TY
yty3805595/YOLO_v3_tutorial_from_scratch
Accompanying code for Paperspace tutorial series "How to Implement YOLO v3 Object Detector from Scratch"