This project aims to extract weather attributes from a given image.

Please download weights from here: https://drive.google.com/drive/folders/1pgHcN1kikHKete80SqmyY2p9cIL3c67t

And test data: https://drive.google.com/drive/folders/1MIxsMe42t6Grb07g6jAmt5SnIhpe4MDR

Run $python infer_attr.py ./path-to-images

Models aim to achieve high accruacy with small models sizes.

Mask pixel error rate: 1.5%

Data: ADE20K

Mask pixel error rate: 10%

Data: SWIMSEG

Mask pixel error rate: 4.6%

Data: SBU-shadow

Accruacy: 93%

Data: PRIVATE

Accruacy: 93%

Data: PRIVATE

Mask pixel error rate: 3.1%

Intensity Accracy: 88%

This program is also migrated to C++ for software integration. However, the code belongs to Hong Kong Applied Science and Technology Research Institute and is sold in the product SRACE.

Copyrights reserved by Liang-yu (Charles) Chen

Should you have any enquiries about the project and dataset, please email lc3533@columbia.edu