The data for the the paper "Atom Cloud Detection and Segmentation Using a Deep Neural Network"

The zip file has three separate directories. 

The first (optical_depths) has the optical depth data for the images in figures 2 and 4.
The second has all of the experimental image data first normalized and then converted to rgb using the viridis colormap.
The third contains a pickle file with the manually labelled data. 
When unpickled this yields a dataframe with the following 10 columns:

1) num_objs-the number of clouds in the image
2) label-list with the cloud labels 1 for MOT and 2 for ODT
3) xmin-list with xmin ROI coordinate in the same order as the list in column 2
4) ymin-list with ymin ROI coordinate in the same order as the list in column 2
5) xmax-list with xmax ROI coordinate in the same order as the list in column 2
6) ymax-list with ymax ROI coordinate in the same order as the list in column 2
7) run-experimental run number
8) type-the image type absorption of fluorescence
9) path-str for png image in directory 2
10) train_val-whether the image is part of the training or validation dataset