Change Project structure to be a little more scalable

[Hennzau]

Objective

Hi! This is the final useful PR for #4. After ensuring everything was easy to use and to create new datatypes, I slightly changed the entire structure of the project to make it a bit more scalable. Here’s how it looks: 😎

Usage

To create a new datatype, you need to add a module in the libraries/datatypes crate. If you want to add ndarray or arrow support for this datatype, you’ll need to create a ndarray.rs or arrow.rs submodule that will use the functions and structures from the libraries/converter crate. For example, for BBox’s arrow support, we use the fastformat_converter::arrow module:

use fastformat_converter::arrow::{FastFormatArrowBuilder, FastFormatArrowRawData};

impl<'a> BBox<'a> {
    pub fn raw_data(array_data: arrow::array::ArrayData) -> Result<FastFormatArrowRawData> {
        use arrow::datatypes::Float32Type;

        let raw_data = FastFormatArrowRawData::new(array_data)?
            .load_primitive::<Float32Type>("data")?
            .load_primitive::<Float32Type>("confidence")?
            .load_utf("label")?
            .load_utf("encoding")?;

        Ok(raw_data)
    }

    pub fn from_raw_data(mut raw_data: FastFormatArrowRawData) -> Result<Self> {
        use arrow::datatypes::Float32Type;

        let data = raw_data.primitive_array::<Float32Type>("data")?;
        let confidence = raw_data.primitive_array::<Float32Type>("confidence")?;
        let label = raw_data.utf8_array("label")?;
        let encoding = Encoding::from_string(raw_data.utf8_singleton("encoding")?)?;

        Ok(Self {
            data: Cow::Owned(data),
            confidence: Cow::Owned(confidence),
            label,
            encoding,
        })
    }
}

We do the same for ndarray support. 🤓

Features

I’ve added the [[features]] tag for the fastformat crates, so that we don’t add a bunch of dependencies for people who don’t need them. 🚀

[dependencies]
fastformat = { version = "0.1.0", features = ["arrow", "ndarray"] }

[haixuanTao]

Relevant to this PR: kornia/kornia-rs#115

I talked with @edgarriba and he is happy to work with us to make kornia a thing in dora-rs. And as performance is a key criteria for dora, it's interesting that if i recall correctly kornia can be slightly faster than opencv.

[edgarriba]

would be great to support a kornia Image/Tensor converter

• image: https://github.com/kornia/kornia-rs/tree/main/crates/kornia-image/src
• tensor: https://github.com/kornia/kornia-rs/tree/main/crates/kornia-image/src

[edgarriba]

you might want to give a shot to the kornia camera streamer:

• rtsp: https://github.com/kornia/kornia-rs/tree/main/examples/rtspcam
• v4l2: https://github.com/kornia/kornia-rs/tree/main/examples/webcam

[edgarriba]

same here, some support conversion from kornia-dnn (to landed this week).
Still iterating on the Detection struct, but for now what we have is sufficient enough to keep it lightweighted

• implement kornia-dnn with RTDETR detector kornia/kornia-rs#129

[edgarriba]

@Hennzau @haixuanTao the kornia-dnn basically is a wrapper around the ort crate which I'm iterating with the author in order to run inference the models from kornia-pytorch-models from the HF hub: https://huggingface.co/kornia/ONNX_models that @shijianjian has been working on

[Hennzau]

Adding Kornia is a really good idea, I made this PR so it's easy to add/change conversion wrapper!

I opened #8 to make all the needed changes :)