v0.0.19

What's Changed

• Add Apple ml-depth-pro model by @jamjamjon in #46

Full Changelog: v0.0.18...v0.0.19

Model	Demo
Depth Anything v1-s (512 x 512)
Depth Anything v2-s (512 x 512)
Depth Pro (1536 x 1536)

v0.0.18

What's Changed

• Add support for restricting detection classes by @jamjamjon in #45

Full Changelog: v0.0.17...v0.0.18

v0.0.17

What's Changed

• Add YOLOv11 by @jamjamjon in #42
• Minor fix by @jamjamjon in #43

Full Changelog: v0.0.16...v0.0.17

v0.0.16

What's Changed

• v0.0.16 by @jamjamjon in #41
  Full Changelog: v0.0.15...v0.0.16

More Detailed

• Add with_device() for Options. Now you can use:

  let options = Options::new().with_device(Device::Cuda(2));
  
  // or
  let options = Options::new().with_cuda(2);

• Add .with_ixx to simplify model building with Options.

  // before
  let options = Options::new()
      .with_i00((1, 2, 4).into())
      .with_i02((416, 640, 800).into())
      .with_i03((416, 640, 800).into());
  
  // now
  let options = Options::new()
    .with_ixx(0, 0, (1, 2, 4).into())
    .with_ixx(0, 2, (416, 640, 800).into())
    .with_ixx(0, 3, (416, 640, 800).into());

  Note: The above only applies to models with dynamic shapes. If your model does not have dynamic shapes, you do not need to call this interface. Additionally, if you are not using TensorRT, you can modify .with_ixx(0, 0, (1, 2, 4).into()) to .with_ixx(0, 0, 2.into()), or you can use .with_batch_size(2) to uniformly set the batch size for all inputs.

• Update MinOptMax struct. Data type changed from isize to usize. You can create it like:

  let default_mom = MinOptMax::default(); // min = opt = max = 1
  let ones_mom = MinOptMax::ones(); // min = opt = max = 1
  let zeros_mom = MinOptMax::zeros(); // min = opt = max = 0
  let from_i32: MinOptMax = MinOptMax::from(-5); // min = opt = max = 0
  let from_i64: MinOptMax = MinOptMax::from(-3); // min = opt = max = 0
  let from_u32: MinOptMax = MinOptMax::from(4u32); // min = opt = max = 4
  let from_u64: MinOptMax = MinOptMax::from(7u64); // min = opt = max = 7
  let from_usize: MinOptMax = MinOptMax::from(10); // min = opt = max = 10
  let from_isize: MinOptMax = MinOptMax::from(-1isize); // min = opt = max = 0
  let from_f32: MinOptMax = MinOptMax::from(-2.0); // min = opt = max = 0
  let from_f64: MinOptMax = MinOptMax::from(3.9); // min = opt = max = 4
  let tuple_mom: MinOptMax = MinOptMax::from((1, 2, 3)); // min = 1, opt = 2, max = 3
  let array_mom: MinOptMax = [1, 2, 3].into(); // min = 1, opt = 2, max = 3

• Add Viewer for image displaying and video saving

let mut viewer = Viewer::new().with_delay(10).with_scale(1.).resizable(true);

// iteration
for (xs, _) in dl {
    // inference & annotate
    let ys = model.run(&xs)?;
    let images_plotted = annotator.plot(&xs, &ys, false)?;

    // show image
    viewer.imshow(&images_plotted)?;

    // check out window and key event
    if !viewer.is_open() || viewer.is_key_pressed(usls::Key::Escape) {
        break;
    }

    // write video
    viewer.write_batch(&images_plotted)?;
}

// finish video write
viewer.finish_write()?;

v0.0.15

What's Changed

• Add florence2-base model by @jamjamjon in #37

Full Changelog: v0.0.14...v0.0.15

Demo

Task	Demo
Caption-To-Phrase-Grounding
Ocr-With-Region
Dense-Region-Caption
Object-Detection
Region-Proposal
Referring-Expression-Segmentation

v0.0.14

What's Changed

• Added Hub for resource management

  <tag/filename> is from release page, e.g. yolo/v5-n-dyn.onnx, images/bus.jpg

  Hub::new()?.fetch(<tag/filename>)?.commit()?;

• Updated DataLoader to support video and streaming

  let dl = DataLoader::new(
          "./assets/bus.jpg", // local image
          // "images/bus.jpg",  // remote image
          // "../images", // image folder
          // "../demo.mp4",   // local video
          // "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4",  // remote video
          // "rtsp://admin:xyz@192.168.2.217:554/h265/ch1/",  // rtsp h264 stream
      )?
      .with_batch(4)
      .with_progress_bar(true)
      .with_bound(100)
      .build()?;
      
      // run
      for (xs, _) in dl {
          // std::thread::sleep(std::time::Duration::from_millis(1000));
          let ys = model.forward(&xs, false)?;
          annotator.annotate(&xs, &ys);
      }

• Updated CI

• Replaced println! with tracing for logging

Full Changelog: v0.0.13...v0.0.14

v0.0.13

What's Changed

• Support auto download ort feature
• Fix polygon drawing

Full Changelog: v0.0.12...v0.0.13

v0.0.12

What's Changed

• Add sapiens-seg model by @jamjamjon in #33

Full Changelog: v0.0.11...v0.0.12

v0.0.11

What's Changed

• ONNXRuntime -> 1.19.0
• CUDA -> 12.x
• TensorRT -> 10.x

Full Changelog: v0.0.10...v0.0.11

v0.0.10

What's Changed

• Add Xs, a wrapper over Vec<x> by @jamjamjon in #29
• Add GroundingDINO by @jamjamjon in #30

Full Changelog: v0.0.8...v0.0.10