Landing Pad Data Cleaning Instructions
Â
Thank you for taking the time to help out with cleaning the training data for the Landing Pad Detection model. We really appreciate your time and effort! Below are the steps you need to follow to successfully complete the task at hand. Good luck!
Overview
On Sunday April 2nd, the Autonomy team collected landing pad footage for ML training for competition. The footage has been sliced into images and has been stored on the team’s OneDrive (2023-04-02 Pilot DO NOT MOVE ). The original videos are there for reference as well.
We want to keep the images with blue landing pads to label for training.
One cell takes about 5 - 10 minutes to complete if you view all the images in tiles.
Task
Keep images with at least 1 whole blue landing pad. Delete all other images.
We do not care about orange landing pads.
If you are unsure if a blue landing pad exists, just delete. Make sure to check the edges of the image!
- There is a blue landing pad in this image, but it might as well not exist. Delete.
- Zoomed in version of the above image. That blob could be anything.
Text and other overlay is fine, as long as most of the landing pad can be seen.
There should be at least 1 non-clipped landing pad.
If the landing pad is blocked by a leg, count it as clipped.
Discolouration or static is fine, as long as the landing pad is still visible.
Instructions
Add your name and subteam to the sign up Excel sheet (in OneDrive→2023-04-02 Pilot DO NOT MOVE→.Signup.xlsx ). Keep in mind that each image should take a few seconds (average of 5s each at most).
The images in each folder are labelled numerically (e.g. 0001.png, 0002.png). Please update the progress column with the image number you are on so that everyone can see how close you are to completing the task.
In the status column please indicate your status (Not started, In progress, Done).
Start deleting images! Go to the OneDrive link above, click on the appropriate folder, and then on the image to display it. To delete an image, click the three dots on the top bar and select the delete button.
There are reference videos as well to help you (they are at 30 FPS). But we need images to be deleted, not which parts of the videos have/lack landing pads.
The hard deadline for this task to be completed is 2359 on Monday, 10 April 2023.
Once again thank you so much for your help with this task. We understand that this task is very tedious and repetitive but your efforts help us get ready to train a model for competition . If you have any questions please feel free to ping @Mihir or @Xierumeng in the #landing-pad channel on discord.
Related Links
Â
Â