-
Notifications
You must be signed in to change notification settings - Fork 32
Open
@Assassyn
Description
Hi,
I am trying to connect FastOpenCV with onnxruntime-react-native, but I am unsure how to do it in JavaScript with FastOpenCV. I have similar functionality implemented in F#, and it looks like that:
`let private normalizeAndResize (image: Mat) =
let targetWidth = 1024
let targetHeight = 1024
let mean = [| 0.485f; 0.456f; 0.406f |]
let std = [| 1.0f; 1.0f; 1.0f |]
// Resize the image
let workingImage = image.Resize(new Size(targetWidth, targetHeight))
// Create tensor for image data
let input = DenseTensor<float32>([| 1; 3; targetHeight; targetWidth |])
for y in 0 .. targetHeight - 1 do
for x in 0 .. targetWidth - 1 do
let pixel = workingImage.Get<Vec3b>(y, x)
// Normalize pixel values
input.[0, 0, y, x] <- (float32 pixel.Item2 / 255.0f - mean.[0]) / std.[0] // R channel
input.[0, 1, y, x] <- (float32 pixel.Item1 / 255.0f - mean.[1]) / std.[1] // G channel
input.[0, 2, y, x] <- (float32 pixel.Item0 / 255.0f - mean.[2]) / std.[2] // B channel
input`
My main problem is that I do not know how to extract pixels from Mat. I have tried to use toJSValue, but for an image 1024x1024 it produces only around 300k of bytes instead of the desired ~1M
Metadata
Metadata
Assignees
Labels
No labels