1. Upload your compressed folder of videos or compressed folder of folders of frames (must be ‘.jpg’, ‘.jpeg’, or ‘.png’) where each video or subfolder corresponds to a labeling job you want to start (even if you only want to start one labeling job you still need to keep this structure so you would just have one subfolder in the outer folder).
2. Create and upload your label job config file. See the user guide for detailed instructions: (https://docs.google.com/document/d/131ZnsRZRiKqgSjmkBkiuRrpHOMlC45UJxQm36pQaf30/edit)
3. Submit the Neurocaas job by pressing the red “submit” button.
4. Wait around 8-10 minutes for AWS to create the labeling jobs. If you are starting multiple jobs with videos you might have to wait longer.
5. Each labeler should be sent an email with their username and temporary password. If they do not see it they should check their spam or junk email.
6. Each labeler can go to the link: pgvx2rzogw.labeling.us-east-1.sagemaker.aws and sign in with their username and temporary password. They will be asked to create a new password that they should keep track of. Please note that if you start another round of labeling jobs at a later date with labelers with the same username and email, they will not be sent another email as they are already part of the system, and all they have to do is go to the above link again and sign in with their credentials.
7. They will be redirected to a list of labeling jobs that have been created, and they can select a labeling job to start working on.
8. Label your data. Here are some guidelines and tips for using the labeling GUI: The most efficient labeling method is to first select one body part, and then label that body part across all frames, using the arrow keys to quickly switch from frame to frame. After all frames are labeled with a single body part, move on to the next body part. You can select a placed label with the mouse and then press the “delete” key to delete it from the frame. You can see all keyboard controls and shortcuts from the “instructions” tab of the labeling GUI on the top left of the screen. You should Remember not to apply more than one label for a specific body part to an individual frame. When you label a body part in an image your label will be given an instance number, but just ignore that.
9. Once you have labeled all the frames you can press the orange “submit” button, but you can also press the “stop and resume later” button on the screen’s top right to return to the task at a later time. Sometimes it will take a few moments to submit, but please don’t close your window until it has completed submitting.
10. Once the labeling job has been completed your labeled data will be submitted to an s3 bucket and formatted so that it is ready to use to train a model. The labeled data can be accessed using the same Neurocaas page you used to trigger the labeling jobs by selecting your labeled dataset file from the results section on the right side of the page. The file will be under the “results/process_results/labeled_data” subdirectory. If you plan on labeling more data to be added to the training dataset that you created as part of this Neurocaas job, please make note of the job ID of this Neurocaas job which is found under the “Status” column. For the next set of labeling jobs you launch from Neurocaas that you want to be combined with your existing labeled data you should then use this Neurocaas job ID as input for the “prevneurocaasjobID” field in the config file.
11. Thanks for using! If you want to access files related to this Neurocaas job in the future after leaving the page you can go to the job history section of the “Labeling for Pose Tracking” analysis page on the Neurocaas website and click on the page corresponding to this job ID.