This demo showcases how to use CLIP to compute similarity between texts and images, which can be employed for image-to-text or text-to-image retrieval tasks.
cargo run -r --example clip
clip-b32-visual
clip-b32-textual
// visual
let options_visual = Options::default()
.with_model("VISUAL_MODEL") // <= modify this
.with_i00((1, 1, 4).into())
.with_profile(false);
// textual
let options_textual = Options::default()
.with_model("TEXTUAL_MODEL") // <= modify this
.with_i00((1, 1, 4).into())
.with_profile(false);
cargo run -r --example clip
(82.24775%) ./examples/clip/images/carrot.jpg => 几个胡萝卜
[0.06708972, 0.0067733657, 0.0019306632, 0.8224775, 0.003044935, 0.083962336, 0.014721389]
(85.56889%) ./examples/clip/images/doll.jpg => There is a doll with red hair and a clock on a table
[0.0786363, 0.0004783095, 0.00060898095, 0.06286741, 0.0006842306, 0.8556889, 0.0010357979]
(90.03625%) ./examples/clip/images/peoples.jpg => Some people holding wine glasses in a restaurant
[0.07473288, 0.0027821448, 0.0075673857, 0.010874652, 0.003041679, 0.0006387719, 0.9003625]
- TensorRT support for textual model