Data is located in /network/tmp1/ccai
data/folder contains the raw datasetspreprocessed/contains data prepared/preprocessed for inference (format, crops, etc.)results/contains the results of training procedures: intermediate/final inferences, checkpoints etc.climate/contains data and notebooks from the climate domain. Data description in climate-code/data.md.
| Folder name | description |
|---|---|
| 27.02.FloodingData | First version of the dataset with open-access images of houses and flooded areas. Includes images of Venice and some donated by a photographer |
| 280_flooded_houses | 280 images of flooded houses segmented by the ccai team with LabelBox. Masks are binary |
| deeplab_segmented_houses | Images of houses from the 1k dataset segmented by deeplab: ground labels are merged together to create a binary mask. Labels merged as ground: Road, sidewalk, terrain. More info in floods-gans/ground_segmentation |
| mapillary | dataset w/ labels, instances and panoptic segmentation https://research.mapillary.com/ |
| SimFlood 50-50f-50p-50m | 50 Synthetic images of flooded/non-flooded pairs + pink images (where the water is) + masks (where the pink was -> see pink_to_mask.py in various scripts) |
| val_set | CCAI's validation set: Currently, the validation set consists of 50 streetview images (random selection of coordinates) but it's made sure that images are from urban/suburban areas and contain atleast one house/building (mostly from urban) |
| video_water_database | Images with masks from Water Detection through Spatio-Temporal Invariant Descriptors, http://isis-data.science.uva.nl/mettes/VideoWaterDatabase.tar.gz |
| elementai_data | Simulated data from Element AI: Contains 10K images without stormy weather in trainA and 10K images with stormy weather in trainB |
| elementai_mapillary | Data for Domain Adaptation : Contains 10K images from Element AI without stormy weather in trainA and 18K images (across seasons) from mapillary dataset in trainB |
| real_to_sim_non-flooded | Data for Real-to-Sim non-flooded style transfer : Contains 1136 real non-flooded images(Cityscapes 982 + streetview 154) in trainA and 10K images from Element AI without stormy weather in trainB |
| /network/tmp1/ccai/preprocessed/munit/depth_floods_280 | Data for MUNIT-LOMIT : Exact details of how the data is being used is mentioned in the config file here. |
| /network/tmp1/ccai/preprocessed/checkpoints_instagan_experiments | Checkpoints for various experiments with InstaGAN on different kinds of datasets : The best results are obtained with checkpoints in folder /depth_flood_280_exps/depth_floods_280_instagan_exp1 |