The goal of the Kinetics dataset is to help the computer vision and machine learning communities advance models for video understanding. Given this large human action classification dataset, it may be possible to learn powerful video representations that transfer to different video tasks.
Avoid technical jargon in the report to keep it accessible. Structure the report with sections like Introduction, Key Points, How to Access, Conclusion. Keep it concise and user-friendly.
Also, consider the user's intent. They might need the document for study, professional use, or compliance. The report should cover the purpose, scope, key sections, and how to legally obtain it. Need to mention any updates or versions released after 2023, but since my knowledge is up to 2023, I can only refer to what was current then.
Wait, maybe they want to access the document but can't find a direct link. I should check if there's an official source. ICAO's website usually has public documents, but some might require purchase or subscription. Also, Doc 9811 is likely a public document since many ICAO standards are open.
Avoid technical jargon in the report to keep it accessible. Structure the report with sections like Introduction, Key Points, How to Access, Conclusion. Keep it concise and user-friendly.
Also, consider the user's intent. They might need the document for study, professional use, or compliance. The report should cover the purpose, scope, key sections, and how to legally obtain it. Need to mention any updates or versions released after 2023, but since my knowledge is up to 2023, I can only refer to what was current then.
Wait, maybe they want to access the document but can't find a direct link. I should check if there's an official source. ICAO's website usually has public documents, but some might require purchase or subscription. Also, Doc 9811 is likely a public document since many ICAO standards are open.
1. Possible to use ImageNet checkpoints?
We allow finetuning from public ImageNet checkpoints for the supervised track -- but a link to the specific checkpoint should be provided with each submission.
2. Possible to use optical flow?
Flow can be used as long as not trained on external datasets, except if they are synthetic.
icao doc 9811 pdf link
3. Can we train on test data without labels (e.g. transductive)?
No.
Avoid technical jargon in the report to keep it accessible
4. Can we use semantic class label information?
Yes, for the supervised track.
Also, consider the user's intent
5. Will there be special tracks for methods using fewer FLOPs / small models or just RGB vs RGB+Audio in the self-supervised track?
We will ask participants to provide the total number of model parameters and the modalities used and plan to create special mentions for those doing well in each setting, but not specific tracks.