Dear participants for GAZE 2021 competition,
We updated our terms and conditions for the competition with the following new terms:
- When performing inference for generating predictions for Codalab, you must perform this sample-by-sample without leaking any information between the samples. Specifically, no known geometry information should be exploited to predict or correct the predictions of more challenging samples, by using predictions from easier samples and their relative camera transformation or head orientation.
- Other test dataset biases and characteristics should not be taken advantage of, such as the use of 18 fixed-position cameras and information implied from the ordering and grouping of the dataset.
- No hand-labeled values should be used as input when performing inference on the test set. The input for the inference of a single sample should be the image itself only.
The purpose of these updates is to keep a fair competition and to avoid any trick that does not benefit the community. Please also note that the winners of the GAZE 2021 competition must open-source their code that allows for the recreation of their final reported performance. Thank you for your understanding.
Kind regards,
ETH-XGaze team