Fixing for tract

examples/wasm-emscripten/Cargo.toml

@@ -9,10 +9,19 @@ path = "src/main.rs"
 name = "wasm_example"
 
 [dependencies]
-ort = { path = "../../", default-features = false, features = ["ndarray", "webgpu", "download-binaries"] }
+ort-tract = "0.1.0+0.21"
+#ort = { path = "../../", default-features = false, features = ["ndarray", "download-binaries"] }
 ndarray = "0.16"
 image = "0.25"
 
 [build-dependencies]
 glob = "0.3"
 reqwest = { version = "0.12", features = ["blocking"] }
+
+[dependencies.ort]
+version = "=2.0.0-rc.10"
+default-features = false # Disables the `download-binaries` feature since we don't need it
+features = [
+    "ndarray",
+    "alternative-backend"
+]

examples/wasm-emscripten/serve.py

@@ -23,6 +23,6 @@ def end_headers (self):
 
 # Serve index.html.
 chdir(path.join(path.dirname(__file__), f"target/wasm32-unknown-emscripten/{mode}"))
-httpd = HTTPServer(("localhost", 5555), CORSRequestHandler)
-print(f"Serving {mode} build at: http://localhost:5555")
+httpd = HTTPServer(("0.0.0.0", 5555), CORSRequestHandler)
+print(f"Serving {mode} build at: http://0.0.0.0:5555")
 httpd.serve_forever()

examples/wasm-emscripten/src/main.rs

@@ -39,23 +39,18 @@ pub extern "C" fn dealloc(ptr: *mut std::os::raw::c_void, size: usize) {
 
 #[no_mangle]
 pub extern "C" fn detect_objects(ptr: *const u8, width: u32, height: u32) {
+	ort::set_api(ort_tract::api());
 	ort::init()
-		.with_global_thread_pool(ort::environment::GlobalThreadPoolOptions::default())
 		.commit()
 		.expect("Cannot initialize ort.");
 
 	let mut builder = ort::session::Session::builder()
 		.expect("Cannot create Session builder.")
 		.with_optimization_level(ort::session::builder::GraphOptimizationLevel::Level3)
-		.expect("Cannot optimize graph.")
-		.with_parallel_execution(true)
-		.expect("Cannot activate parallel execution.")
-		.with_intra_threads(2)
-		.expect("Cannot set intra thread count.")
-		.with_inter_threads(1)
-		.expect("Cannot set inter thread count.");
+		.expect("Cannot optimize graph.");
 
-	let use_webgpu = true; // TODO: Make `use_webgpu` a parameter of `detect_objects`? Or say in README to change it here.
+
+	let use_webgpu = false; // TODO: Make `use_webgpu` a parameter of `detect_objects`? Or say in README to change it here.
 	if use_webgpu {
 		use ort::execution_providers::ExecutionProvider;
 		let ep = ort::execution_providers::WebGPUExecutionProvider::default();
@@ -66,10 +61,14 @@ pub extern "C" fn detect_objects(ptr: *const u8, width: u32, height: u32) {
 		}
 	}
 
+	let time_start = std::time::Instant::now();
 	let mut session = builder
 		.commit_from_memory(include_bytes!("../yolov8m.onnx"))
 		.expect("Cannot commit model.");
+	println!("Model loaded in {} ms", time_start.elapsed().as_millis());
+
 
+	let time_start = std::time::Instant::now();
 	let image_data = unsafe { std::slice::from_raw_parts(ptr, (width * height * 4) as usize).to_vec() }; // Copy via .to_vec might be not necessary as memory lives long enough.
 	let image = image::ImageBuffer::<image::Rgba<u8>, Vec<u8>>::from_raw(width, height, image_data).expect("Cannot parse input image.");
 	let image640 = image::imageops::resize(&image, 640, 640, image::imageops::FilterType::CatmullRom);
@@ -79,6 +78,7 @@ pub extern "C" fn detect_objects(ptr: *const u8, width: u32, height: u32) {
 
 	let outputs: ort::session::SessionOutputs = session.run(ort::inputs!["images" => tensor]).unwrap();
 	let output = outputs["output0"].try_extract_array::<f32>().unwrap().t().into_owned();
+	println!("time to infer in {} ms", time_start.elapsed().as_millis());
 
 	let mut boxes = Vec::new();
 	let output = output.slice(ndarray::s![.., .., 0]);