Mini-ImageNet Tools: Generation Scripts for Few-Shot Learning Benchmarks

Mini-ImageNet Tools provides scripts for generating the 100-class mini-ImageNet benchmark from raw ImageNet files, authored by yaoyao-liu in 2020. It automates the creation of standardized image batches for meta-learning and few-shot classification research.

Use Cases

• Generating N-way K-shot episodes for meta-learning using the provided batch processing scripts
• Standardizing ImageNet subsets into the 100-class mini-ImageNet format for benchmarking
• Evaluating few-shot classification accuracy on the provided train/val/test split files

Strengths

• MIT licensed for open research and commercial use
• Provides standardized train/val/test splits for 100 classes
• Includes batch processing scripts specifically designed for meta-learning episodes

Limitations

• Requires external 150GB+ ImageNet source data which is not provided
• Temporal staleness with the last update occurring in 2020
• Limited to the specific 100-class mini-ImageNet configuration

Provenance

Source

Derived from the ILSVRC 2012 (ImageNet) dataset

Collection Method

Script-based extraction and subsetting of existing image data

Freshness

Last updated October 2020; no recent maintenance or updates to the generation logic recorded.

Geography

Global