hi there, i have trained some data for a project. now i want to add new data that i collected but only train/valid/test on this subset ingoring the previuos one. how do i do that?
Hello! To add new data and train only on this subset, you can generate a new dataset version and use the “Filter by Tag” feature in Roboflow to only include your new images. Here are the steps:
- First, you need to search your dataset for the new images you’ve collected.
- Once you’ve found them, select these images and tag them.
- After tagging, generate a version of this dataset. Use the “Filter by Tag” transformation to filter which images should be included in a version based on which Tags are applied. You can set the filter to “Require” so that only images with the tags you applied will be included in the version.
- Once you have created a new version with the new data, you can then use this dataset version to train a model in Roboflow. Remember, the new split will be reflected in future version exports. Past dataset versions do not change, as they are to be a source of record.
For more information, you can refer to the Roboflow documentation on creating a dataset version.
However, while this is technically true you generally shouldn’t do this. Your model will succumb to forgetting if you remove past images. I’d recommend training again with the full dataset but from the previous checkpoint.
thanks for sharing your answer, you taught me something new on machine learning:)