DeepSense Forum
Hi all,
thank you for creating this interesting challenge. As I looked at the test dataset and the train dataset the effetive step size is different, e.g.
Original Scenario 34:
1,./scenario34/unit1/camera_data/image_3015.jpg,./scenario34/unit1/camera_data/image_3017.jpg,./scenario34/unit1/camera_data/image_3019.jpg,.
--> 2 Step Size
Test Dataset Scenario 34:
1,./scenario34/unit1/camera_data/image_2006.jpg,./scenario34/unit1/camera_data/image_2007.jpg,./scenario34/unit1/camera_data/image_2008.jpg,.
--> 1 Step Size
My question would be: Is this on purpose and the new dataset does not contain the intermediate steps, e.g. the effective timings are still correct?
BR
Max
Hi Max,
Your observation is correct. Since the main objective of this multi-modal beam prediction challenge is the generalization and fast adaptation to unseen data, the test set was constructed with two main differences compared to the development set: (i) 50% of the test set is constructed from DeepSense Scenario 31 which hasn't been seen in the development dataset, and (ii) the sequences in the test set have a different sampling rate compared to the development set (in particular, the sampling time in the test set sequences is half the sampling time of the development set sequences).
In order to further facilitate the generalization and adaptation task in this competition, we are providing a small adaptation dataset comprising the following:
- 50 labeled samples from the unseen Scenario 31
- 25 labeled samples from Scenario 32
- 25 labeled samples from Scenario 33
The sequences in this adaptation dataset have the same sampling rate as the competition test set.
The adaptation dataset can be downloaded from the same page with the test set download links.
Please let us know if you have any further questions.
Looking forward to your submission!
Regards,
Ahmed
Thank you very much for the fast response.
As I am currently constructing the last part of my solution, and since you shared the full dataset: I could recreate a training dataset with the same step size as the test set.
Is this allowed or should I keep to step size 2?
BR
Max
Hi Max,
I wanted to clarify this point: What we shared a few days ago was not a "full dataset"; it is just a very small (100 data points) adaptation dataset that is meant to augment (and not replace) the original training dataset. You are free however to do anything with this adaptation dataset (e.g., merge it with the original training dataset, use it to refine the model trained on the original training dataset, use it instead of the training set, neglect it, use it to test your model, or any other ideas you have).
I hope this helps.
Regards,
Ahmed
