Mentioned directory format in the README is missing 'train' , 'test' and 'hold' directories, process_data.py about xview2_baseline HOT 7 CLOSED

Thank you for pointing this out! This will be added to the README but process_data.py assumes you have an output directory with three empty subfolders for train, test, and hold when extracting the polygons. However, since the challenge provides you with separate train and test downloads, you can modify the script for lines 165 - 175. from what is below to the second code snippet. Then you can run it on the train and test dataset directories independently and structure the directories like https://github.com/DIUx-xView/xview2-baseline#other-methods

In addition, please change line 148 to be the CSV name of your choice:
df.to_csv(INSERTNAME.csv)

Before Modification

    logging.info("Started Processing for Train Data")
    process_data(args.input_dir, args.output_dir, 'train')
    logging.info("Finished Processing Train Data")

    logging.info("Started Processing Test Data")
    process_data(args.input_dir, args.output_dir, 'test')
    logging.info("Finished Processing Test Data")

    logging.info("Started Processing Holdout Data")
    process_data(args.input_dir, args.output_dir, 'hold')
    logging.info("Finished Processing Holdout Data")

After Modification

    logging.info("Started Processing for Data")
    process_data(args.input_dir, args.output_dir, '')
    logging.info("Finished Processing Train Data")

from xview2_baseline.

Thanks for replying @ssajeev .
I understand output folder will have train, test and hold dirs. My doubt was regarding the input directory. I suppose your modification will solve the issue above.

However this creates a subsequent issue. Addressing it here:
After finishing the process_data.py step, next step is to run damage_classification.py script that also takes --test_data and --test_csv as input. Looking at the script, this set has to be the validation set.
Now to obtain this validation set, I feel process_data.py needs changes to rng output train.csv and test.csv for the same input train directory.

Hopefully I made myself clear.

from xview2_baseline.

For now, we recommend splitting the train data beforehand rather than doing it in process_data.py. We will provide an updated process_data.py script that will split the data in train into train and validation. We will update the README to reflect this.

from xview2_baseline.

Great!
I would prefer not to close the issue before it's actually addressed in the code.
Thanks again.

from xview2_baseline.

The code and README have been updated with new instructions for process_data.py. Please verify that it works for you.

from xview2_baseline.

@ice-pice Thanks for putting in this issue. We are going to mark it resolved. If you see any other problems please open a new issue, thanks again!

from xview2_baseline.

Hello, where did you get the label in the test dataset? The test data set I downloaded has no labels.

from xview2_baseline.