Illustration2Vec

Provides downloadable pre-trained CNN models for immediate use without training, as shown in the tag prediction and feature extraction examples in the README.

Supports both Caffe and Chainer frameworks, allowing users to choose based on existing setups, mentioned in the requirements and initialization functions.

Estimates tags across four detailed categories—general, copyright, character, and rating—providing rich semantic insights, evidenced by the output from estimate_plausible_tags().

Extracts both 4096-dimensional real vectors and compact 4096-bit binary vectors, enabling flexible storage and similarity search for applications like image databases.