vpulab / Semantic-Aware-Scene-Recognition

Code repository for paper https://www.sciencedirect.com/science/article/pii/S0031320320300613 @ Pattern Recognition 2020
MIT License
100 stars 16 forks source link

Places-365 noisy training data not found #9

Open KriteshGarg opened 4 years ago

KriteshGarg commented 4 years ago

Tried bash ./Scripts/download_Places365_extra.sh ./Data/Datasets/places365_standard command to get the precomputed Semantic Segmentation masks.

Training masks are not downloaded only it contains validation masks.

Please, update/fix the sh file to provide training data for precomputed Semantic Segmentation masks.

Thanks.

alexlopezcifuentes commented 4 years ago

From now we have only uploaded training Semantic Segmentation masks for ADE, SUN and MIT. Places 365 training data is much heavier and we are still on the process of uploading.

However, you can use any of the semantic segmentation networks from the state-of-the-art to train the model. You just need to meet the format requirements by getting images containing the Top@3 predictions for each pixel.