the links in the data section are for the 16 chunks of training data, training labels and metadata and sample submission, Where is the link to the test data?
@Daz_Mashaly Hi Daz, using the usual terminology for ML competitions, the “test” data is private and not available to competitors.
In this context, the training data means it’s all fair game for you to train your models with, which will almost certainly involve a further train/test split. When your code runs, it will be making predictions on the (private) test data.
Hope that clears up any confusion from overloading the term “test” data.
I think he meant that how do we make a submission when there is no specified test dataset ?
In this competition, you submit code that runs remotely to perform inference on the test dataset. The test dataset is not provided to you.
Documentation about the test set and about the code execution are here:
- Detection Track “Public vs private datasets” / “Code submission format”
- Pose Estimation Track “Public vs private datasets” / “Code submission format”
Hi,
The sample submission and the training data have the same ‘image_id’. Normally we select some portion of the training data to test our predictions before applying it to the private dataset. So I’m curious how to use a sample file to fill our predictions before submission. Just like in another competition forum, where they provide some part of the private data, and during submission, they run the full private test data against the inference code.
Hi @hark99,
This section of the benchmark blog post walks you through how to select a portion of the training data for local testing. I hope this helps.
Mike