Implementation:Ggml org Ggml Sam image preprocess
Sam Image Preprocess
C++ implementation of image preprocessing for the Segment Anything Model (SAM) in the GGML framework.
Primary API
sam_image_preprocess
| Attribute | Value |
|---|---|
| Signature | bool sam_image_preprocess(const sam_image_u8 & img, sam_image_f32 & res)
|
| Source | examples/sam/sam.cpp:L443-505
|
| Repository | https://github.com/ggml-org/ggml |
| Language | C++ |
| Parameters | img (input uint8 image), res (output float32 preprocessed image)
|
| Returns | bool success; populates res with a 1024x1024 float32 image (bilinear-resized, ImageNet-normalized, zero-padded)
|
Processing steps performed:
- Bilinear resize to 1024x1024 with aspect-ratio preservation
- ImageNet normalization (mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375])
- Zero-padding of remaining area after resize
sam_image_load_from_file
| Attribute | Value |
|---|---|
| Signature | bool sam_image_load_from_file(const std::string & fname, sam_image_u8 & img)
|
| Source | examples/sam/sam.cpp:L417-433
|
| Purpose | Loads a JPEG/PNG image from disk into the sam_image_u8 struct using stb_image
|
YOLO Equivalents
The YOLO example provides analogous preprocessing functions with a different pipeline:
load_image
| Attribute | Value |
|---|---|
| Source | examples/yolo/yolo-image.cpp:L66-89
|
| Purpose | Loads and decodes an image from file into a YOLO-compatible image struct |
letterbox_image
| Attribute | Value |
|---|---|
| Source | examples/yolo/yolo-image.cpp:L144-160
|
| Purpose | Resizes image to 416x416 using letterbox strategy with 0.5 gray padding, preserving aspect ratio; outputs CHW planar layout |
Dependencies
- stb_image.h -- Single-header library used for JPEG/PNG image decoding
- Standard C++ math -- Used for bilinear interpolation and normalization arithmetic
Related
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment