Key Considerations Before Submitting Your Algorithm Docker

Hello, participants who are reading this message,

This challenge requires the submission of an algorithm's docker with specific requirements. There are several important points you must be aware of to ensure successful participation.

• Environment Configuration: The algorithm docker will run on Amazon's ml.m5.large instance, equipped with a maximum GPU memory of 32GB, 2 CPU cores, and 8GB of RAM. The allowed processing time for 50 test images is limited to 10 minutes, including model loading, I/O, preprocessing, and inference (for a single stacked image).

• Data Scale: The validation set consists of a stacked MHA file containing 50 images, while the test set comprises three separate MHA files, each containing 50 lateral cephalometric X-ray images.

• Submission Opportunities: During the validation phase, participants have five submission opportunities, and during the testing phase, there are four submission chances. It has been decided to finalize these submission counts, ensuring fair participation in the final round.

• Algorithm Docker Runtime Limit: In the validation phase, the docker will automatically process one mha file, while in the testing phase, it will handle three mha files. Each docker's processing time is limited to 10 minutes, during which it should complete the analysis of 50 images and produce the final landmark predictions.

• Accelerating Inference Speed: A considerable number of submission failures are attributed to the inference time exceeding 10 minutes for processing 50 images, resulting in "Time limit exceeded" errors. To avoid this issue, you can configure the GPU usage to 32GB and try to increase the batch size during the testing phase.

• Ensuring Valid Docker Submissions: To verify the correctness of your constructed docker, you can refer to THIS POST for guidance on how to perform self-checks.

Keeping these essential points in mind will help you prepare and submit your algorithm docker successfully. We wish you the best of luck in the challenge!

Hi jeremyzhang,

Considering that the time & resource requirements were not disclosed from the beginning, would it be possible that we send a docker image and you run it locally?

Best regards, Lisa

Hello, Lisak,

I apologize for any inconvenience caused. The algorithm docker requires participants to upload it using their competition accounts for the results to be valid.

I noticed your previous submission in the backend, and it failed due to a "Time limit exceeded" error, indicating that your code logic is likely correct.

My recommendation is to set the GPU usage to 32GB and try to increase the batch size as much as possible during testing to avoid encountering this time limit error.

For example, since one docker instance will process an MHA file containing 50 stacked images, you can set the testing batch size to 10, allowing for five testing iterations. This might help you bypass the time limit error. (Oh, don't forget to set the model to evaluation mode as well.)

I hope this advice helps you, and please don't hesitate to reach out if you have any further questions.