openvino-mobilenet-image: fix tensor shape

[yamt]

tested with:

• wasmtime
• wasm-micro-runtime with wasi_nn_openvino.c: remove the tensor layout adjustment logic bytecodealliance/wasm-micro-runtime#4308

openvino-mobilenet-raw has the same issue. it can be fixed by transposing the tensor file as suggested in:
as described in
bytecodealliance/wasmtime#10867 (comment)

i guess these tests have the same origin as wasmtime's classification-example and share the same bug.
see
bytecodealliance/wasmtime#10867 for details of the bug.

[juntao]

Hello, I am a code review agent on flows.network. Here are my reviews of changed source code files in this PR.

---

openvino-mobilenet-image/rust/src/imagenet_classes.rs

Potential issues

1. Incorrect String Literal: The string "devil\"s darning needle" contains an escaped double quote (") which is unnecessary and incorrect in Rust string literals. Use a raw string or escape the backslash properly: r#"devil's darning needle"# or "devil's darning needle".

2. Array Length Mismatch: The array IMAGENET_CLASSES is declared with a fixed size of 1000 elements ([&str; 1000]). However, the number of entries provided is less than 1000, which will cause a compilation error. Ensure that all 1000 classes are included or adjust the array's declaration to match the actual number of entries.

3. Lack of Comments: While not a coding issue per se, the absence of comments or documentation around the IMAGENET_CLASSES array makes it harder for future developers to understand its purpose and usage. Consider adding a comment explaining what this array represents.

Summary of changes

• • Removed newline issue: The patch corrects a missing newline at the end of the file, which is now properly terminated.

• No structural code changes: No actual tensor shape or logical changes were made; only formatting was adjusted.

openvino-mobilenet-image/rust/src/main.rs

Potential issues

1. The image::io::Reader::open function might not be available in the environment you're working on.
2. The DynamicImage::resize_exact method requires the dimensions to be a multiple of 32, which is a requirement for some OpenVINO models. If these dimensions are not multiples of 32, the resize operation will fail.
3. The image_to_tensor function assumes that the image pixels are in BGR order, but it does not handle the case where the input image might have different color spaces.

Summary of changes

• Fixed tensor shape for openvino-mobilenet-image.
• Transposed BGR image from [height, width, 3] to [3, height, width].
• Added padding if the input dimensions do not match the expected dimensions.

[hydai]

Thanks

[yamt]

@hydai
can you or someone take care of the other example (openvino-mobilenet-raw) and wasmedge?
i have not used wasi-nn with wasmedge.

[hydai]

Sure, we will track this issue at WasmEdge/WasmEdge#4142.

[yamt]

Sure, we will track this issue at WasmEdge/WasmEdge#4142.

thank you.

[yamt]

openvino-mobilenet-raw has the same issue.

openvino-road-segmentation-adas too.

[hydai]

Thanks, @yamt, for the reminder. I forgot to keep track of these two examples.
Created #192 for tracking them.