Uploading Classification Datasets
If you are looking to label entire images as belonging to a class, then you will need to upload a classification dataset. Classification datasets require images/videos and are in distinct folders. Class names are derived based on the folder names.
- For example, if you are uploading images of dogs, cats, and raccoons, you should have three folders.
- One folder should be called "dogs" and contain all dog images.
- Another folder should be called "cats" and contain all cat images.
- The third folder should be called "raccoons" and contain all raccoon images.
- For detection of a single item in a Classification dataset, select Single-Label Classification
- For detection of multiple items in a Classification dataset, select Multi-Label Classification
Note that if the goal is to detect movement or position, it is better to create an object detection project with the desired data. A Classification project merely answers the question "Is this label or annotation present in this image?"
Want to convert a dataset for use in a classification project?
Converting an Object Detection dataset to a Classification Dataset using Roboflow's "Isolate Objects" Advanced Pre-Processing feature.
Have some data in a Roboflow object detection project that you want to use in a classification project? Here's how:
Generate a dataset in the source project with the Isolate Objects preprocessing step, but no augmentation options, selected.
- This will create crops of all the bounding boxes from the project's dataset for a classifier to be trained on
- Export the data from that version using the OpenAI Clip Classification method (see screenshot) and download to your local machine.
We have built out notebooks and tutorials for fitting five classification models:
There are thousands of image classification datasets publicly available on Roboflow Universe for download and use in your projects.
We regularly write and share blog posts, including tutorials, recommendations for preprocessing and augmenting data, and techniques for improving model performance. Check out our computer vision modeling-related blog posts here and all blog posts here.