Sign in to view source links and access this dataset
Description
Imagenet W21 Webp Wds is a copy of the full Winter21 release of ImageNet in WebDataset tar format with WEBP encoded images. The dataset consists of 19,167 classes, which is 2,674 fewer classes than the original Fall11 release. It was created by author 'timm' and last updated on Hugging Face on 2024-01-07.
Use Cases
Train large-scale image classification models based on the 19,167 class hierarchy.
Benchmark model performance on a standardized, curated subset of ImageNet.
Utilize compressed WEBP images for experiments where storage or bandwidth is a constraint.
Strengths
Contains 19,167 distinct image classes.
Images are encoded in WEBP format, described as being at ~56% of the size of the original.
Classes were reviewed and removed based on documented concerns from the ImageNet team.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Provenance
Source
ImageNet Winter21 release, processed by author 'timm'.
Collection Method
Derived from the original ImageNet release, converted to WebDataset tar format with WEBP encoding.
Freshness
Last updated 2024-01-07 18:13:46; freshness should be verified.
License is unknown and must be verified before use.