search

YXu556 / SITS-MoCo

24 stars 2 forks source link

Require clarity on CSV building logic #3

Open Divyanshu0110 opened 1 month ago

Divyanshu0110 commented 1 month ago

Hello, We are attempting to fine tune the SITS-MoCo model on our dataset which is in .tif files, and we are stuck at making the CSVs properly. What it looks like right now - from the toy data that was uploaded - each sample has 2 dimensions, say (70, 12) where 12 is the number of bands and 70 are the different time steps. How was each image compressed like this? What is the logic here?

YXu556 commented 1 month ago

Hi, thank you for your question.

We provide the code to compress the images in preprocess/2. extract_csv.py. This script stores each pixel as a 2D csv file, and also generates an index csv that stores information about each pixel and locates its corresponding csv file. Then, during model training, the dataset class in datasets/uscrops.py can load all the data through the index file.