huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 03:54:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add DINOv2Reg4 + PlantCLEF2024 (#2293)

Browse Source 
* Add: DINOv2Reg4 with PlantCLEF2024 weights and example ( See https://arxiv.org/abs/2309.16588 and https://zenodo.org/records/10848263 )

* Remove extra files + update README to download them + remove extra lines

* minor fix (README remove extra spaces)

* minor fix (README: Fix image url)

* Modif: Add back interpolate_pos_encoding() + fix when no interpolation + remove extra comments + Update README ( source image changed and so the predictions )

* Fix: Improve code lisibility with '$ cargo clippy' and '$ cargo fmt'

* Another clippy fix.

---------

Co-authored-by: x-VEspit <vincent.espitalier@cirad.fr>
Co-authored-by: laurent <laurent.mazare@gmail.com>
This commit is contained in:
v-espitalier
2024-06-29 11:49:15 +02:00
committed by GitHub
parent a3dd87f15e
commit e27aac0a06
5 changed files with 395 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
candle-examples/src/imagenet.rs
View File

@ -17,6 +17,24 @@ pub fn load_image224<P: AsRef<std::path::Path>>(p: P) -> Result<Tensor> {
.broadcast_div(&std)
}
/// Loads an image from disk using the image crate, this returns a tensor with shape
/// (3, 518, 518). imagenet normalization is applied.
/// The model dinov2 reg4 analyzes images with dimensions 3x518x518 (resulting in 37x37 transformer tokens).
pub fn load_image518<P: AsRef<std::path::Path>>(p: P) -> Result<Tensor> {
let img = image::io::Reader::open(p)?
.decode()
.map_err(candle::Error::wrap)?
.resize_to_fill(518, 518, image::imageops::FilterType::Triangle);
let img = img.to_rgb8();
let data = img.into_raw();
let data = Tensor::from_vec(data, (518, 518, 3), &Device::Cpu)?.permute((2, 0, 1))?;
let mean = Tensor::new(&[0.485f32, 0.456, 0.406], &Device::Cpu)?.reshape((3, 1, 1))?;
let std = Tensor::new(&[0.229f32, 0.224, 0.225], &Device::Cpu)?.reshape((3, 1, 1))?;
(data.to_dtype(candle::DType::F32)? / 255.)?
.broadcast_sub(&mean)?
.broadcast_div(&std)
}
pub const CLASS_COUNT: i64 = 1000;
pub const CLASSES: [&str; 1000] = [