pykeio / ort

hi! i'm working on an example integrating the crate cudarc to perform some image preprocessing using cuda kernels (resize, crop, normalize, etc), which i want to pipeline later to a ort session.

I've seen IoBinding but from the docs and examples i couldn't see any clear direction to achieve that. FYI, this the project i'm working on right now, which potentially i might adopt ort as dnn inference engine for cv tasks.

**https://github.com/kornia/kornia-rs/blob/cuda/examples/cuda_simple/src/main.rs

With 8383879:

ort/examples/cudarc/src/main.rs

Lines 35 to 43 in af95987

    
           let tensor: TensorRefMut<'_, f32> = unsafe { 
        
           	TensorRefMut::from_raw( 
        
           		MemoryInfo::new(AllocationDevice::CUDA, 0, AllocatorType::Device, MemoryType::Default)?, 
        
           		(*device_data.device_ptr() as usize as *mut ()).cast(), 
        
           		vec![1, 3, 512, 512] 
        
           	) 
        
           	.unwrap() 
        
           }; 
        
           let outputs = model.run([tensor.into()])?;

	let tensor: TensorRefMut<'_, f32> = unsafe {
	TensorRefMut::from_raw(
	MemoryInfo::new(AllocationDevice::CUDA, 0, AllocatorType::Device, MemoryType::Default)?,
	(device_data.device_ptr() as usize as mut ()).cast(),
	vec![1, 3, 512, 512]
	)
	.unwrap()
	};
	let outputs = model.run([tensor.into()])?;

conversion from existing cuda slice to ort::Value