Datasets for the benchmark are first submitted to the scratch bucket where they are saved under the "Train" or "Test" folder. Train datasets will be copied fully to the the benchmark bucket. For test datasets the following two changes need to be made:
- framelabels.json will not be included in the copy to the benchmark bucket.
- For cliplabels.json: the first frame's label will be extracted and saved as a start_label.json. start_label.json will be copied to the benchmark bucket. cliplabels.json will not be included in the copy.
Datasets for the benchmark are first submitted to the scratch bucket where they are saved under the "Train" or "Test" folder. Train datasets will be copied fully to the the benchmark bucket. For test datasets the following two changes need to be made: