100 randomly selected classes from the ImageNet-1k collection are provided in this dataset, with all images pre-resized to 160 pixels on the shorter side. Each instance consists of an RGB image and its corresponding integer class label, derived from the original ILSVRC hierarchy.
Use Cases
- Benchmark computer vision models using the 'image' and 'label' columns for classification tasks on a reduced-scale dataset
- Evaluate self-supervised learning algorithms using the provided 100-class subset to minimize training time
- Prototype image processing pipelines using the pre-resized 160px images to speed up data loading and training iterations
Strengths
- Contains 100 distinct object classes sampled from the ImageNet-1k (ILSVRC) dataset
- Images are standardized to a 160-pixel resolution on the shorter side to reduce storage and compute requirements
- Data instances include an 'image' field in RGB format and a 'label' field representing the class index
- Derived from the methodology used in the Contrastive Multiview Coding (CMC) research paper