Function sam3_preprocess_image

sam3.cpp:3490–3518

Preprocess an image: resize to img_size × img_size, convert to float, normalize. Returns a float tensor in [C, H, W] layout (channel-first), range normalized with mean=0.5, std=0.5 → pixel values in [-1, 1].

// Returns a float tensor in [C, H, W] layout (channel-first), range normalized with
// mean=0.5, std=0.5 → pixel values in [-1, 1].
static std::vector<float> sam3_preprocess_image(const sam3_image& image, int img_size) {
    const int C = 3;
    std::vector<float> result(C * img_size * img_size);

    // Resize to img_size × img_size via uint8 bilinear (matching torch pipeline)
    std::vector<uint8_t> resized;
    const uint8_t* pixels = image.data.data();
    int w = image.width, h = image.height;

    if (w != img_size || h != img_size) {
        resized.resize(img_size * img_size * 3);
        sam3_resize_bilinear(pixels, w, h, resized.data(), img_size, img_size);
        pixels = resized.data();
        w = img_size;
        h = img_size;
    }

    // Convert to float [C, H, W] with normalization: (pixel / 255.0 - 0.5) / 0.5
    for (int c = 0; c < C; ++c) {
        for (int y = 0; y < img_size; ++y) {
            for (int x = 0; x < img_size; ++x) {
                float v = pixels[(y * img_size + x) * 3 + c] / 255.0f;
                result[c * img_size * img_size + y * img_size + x] = (v - 0.5f) / 0.5f;
            }
        }
    }

    return result;
}

// SAM2 preprocessing: resize + ImageNet normalization.
// Returns [C, H, W] float, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225].