MIR-FLICKR25K

Click to add a brief description of the dataset (Markdown and LaTeX enabled).

Provide:

Data Documentation

This dataset is a classic multi-label collection from Flickr featuring 25,000 images with text descriptions and tag annotations, widely used for image classification, multi-label classification, and cross-modal retrieval. The package contains two main files: mirflickr25k (with 25,000 images named im*.jpg) and mirflickr25k_annotations_v080 (containing 24 category labels). The text descriptions are preprocessed and include 1386 tags that appear in at least 20 images. Most research filters out zero vectors, resulting in 20,015 usable samples. Traditional approaches extract features using pre-trained models like VGG19 for images (4096-D) and BOW or TextCNN for text (1386-D or 300-D), while modern methods often employ Transformer architectures for downstream tasks. * a high-level explanation of the dataset characteristics * explain motivations and summary of its content * potential use cases of the dataset