test: download test fixtures differently (#140)

* Retrieve OpenVINO test artifacts from https://download.01.org/
This change is a necessary precursor for removing the Git LFS support in
this repository and properly caching these artifacts in GitHub actions.

.github/workflows/main.yml

@@ -39,9 +39,13 @@ jobs:
     - uses: actions/checkout@v2
       with:
         submodules: true
-        lfs: true
-    - name: Checkout LFS objects
-      run: git lfs checkout
+    - uses: actions/[email protected]
+      with:
+        key: openvino-test-fixtures
+        path: |
+          crates/openvino/tests/fixtures/alexnet
+          crates/openvino/tests/fixtures/inception
+          crates/openvino/tests/fixtures/mobilenet
     - uses: abrown/install-openvino-action@v8
       with:
         version: ${{ matrix.version }}
@@ -92,7 +96,6 @@ jobs:
     - uses: actions/checkout@v2
       with:
         submodules: true
-        lfs: true
     - name: Build documentation
       run: cargo doc --no-deps --features openvino-sys/runtime-linking
 

README.md

@@ -20,32 +20,24 @@ crate (high-level, ergonomic bindings) for accessing OpenVINO™ functionality i
 1. The [openvino-sys] crate creates bindings to the OpenVINO™ C API using `bindgen`; this requires a
    local installation of `libclang`. Also, be sure to retrieve all Git submodules.
 
-2. This repo currently uses [git-lfs](https://git-lfs.github.com/) for large file storage. If you
-   [install it](https://github.com/git-lfs/git-lfs/wiki/Installation) before cloning this
-   repository, it should have downloaded all large files. To check this, verify that `find
-   crates/openvino -name *.bin | xargs ls -lhr` returns `.bin` files of tens of megabytes. If not,
-   download the large files with:
-
-   ```shell
-   git lfs fetch
-   git lfs checkout
-   ```
-
-3. This library binds to OpenVINO™'s shared libraries; how those native libraries are configured and
+2. This library binds to OpenVINO™'s shared libraries; how those native libraries are configured and
    installed on your system determines how these Rust bindings work. The [openvino-finder] crate
    attempts to locate the necessary libraries and configuration; if you run into problems, you may
    need to understand additional details documented in the [`openvino-finder`
    docs][openvino-finder-docs].
 
 [openvino-finder-docs]: https://docs.rs/openvino-finder
 
-4. __For macOS (homebrew) users__. Install the openvino toolkit, which includes the native C library,
-   and set `DYLD_LIBRARY_PATH`:
+3. During testing only, this library will download several models for its integration tests. This
+   relies on `curl` being available on the system path.
+
+4. __For macOS (homebrew) users__. Install the openvino toolkit, which includes the native C
+   library, and set `DYLD_LIBRARY_PATH`:
    ```
-       brew install openvino
-       export DYLD_LIBRARY_PATH="$(brew --prefix)/lib"
+   brew install openvino
+   export DYLD_LIBRARY_PATH="$(brew --prefix)/lib"
    ```
-   Then you can build and run openvino-rs for [runtime linking](#build-for-runtime-linking).
+   Then you can build and run using [runtime linking](#build-for-runtime-linking).
 
 ### Build from an OpenVINO™ installation
 

crates/openvino/tests/classify-alexnet.rs

@@ -1,10 +1,11 @@
-//! Demonstrates using `openvino-rs` to classify an image using an AlexNet model and a prepared input tensor. See
-//! [README](fixtures/alexnet/README.md) for details on how this test fixture was prepared.
+//! Demonstrates using `openvino` to classify an image using an AlexNet model and a prepared input
+//! tensor.
+
 mod fixtures;
 mod util;
 
 use anyhow::Ok;
-use fixtures::alexnet::Fixture;
+use fixtures::alexnet as fixture;
 use openvino::{
     prepostprocess, Core, DeviceType, ElementType, Layout, ResizeAlgorithm, Shape, Tensor,
 };
@@ -15,16 +16,16 @@ use util::{Prediction, Predictions};
 fn classify_alexnet() -> anyhow::Result<()> {
     let mut core = Core::new()?;
     let mut model = core.read_model_from_file(
-        &Fixture::graph().to_string_lossy(),
-        &Fixture::weights().to_string_lossy(),
+        &fixture::graph().to_string_lossy(),
+        &fixture::weights().to_string_lossy(),
     )?;
 
     let output_port = model.get_output_by_index(0)?;
     assert_eq!(output_port.get_name()?, "prob");
     assert_eq!(model.get_input_by_index(0)?.get_name()?, "data");
 
     // Retrieve the tensor from the test fixtures.
-    let data = fs::read(Fixture::tensor())?;
+    let data = fs::read(fixture::tensor())?;
     let input_shape = Shape::new(&[1, 227, 227, 3])?;
     let element_type = ElementType::F32;
     let mut tensor = Tensor::new(element_type, &input_shape)?;

crates/openvino/tests/classify-inception.rs

@@ -1,10 +1,11 @@
-//! Demonstrates using `openvino-rs` to classify an image using an Inception SSD model and a prepared input tensor. See
-//! [README](fixtures/inception/README.md) for details on how this test fixture was prepared.
+//! Demonstrates using `openvino` to classify an image using an Inception SSD model and a prepared
+//! input tensor.
+
 mod fixtures;
 mod util;
 
 use anyhow::Ok;
-use fixtures::inception::Fixture;
+use fixtures::inception as fixture;
 use openvino::{
     prepostprocess, Core, DeviceType, ElementType, Layout, ResizeAlgorithm, Shape, Tensor,
 };
@@ -15,16 +16,16 @@ use util::{Prediction, Predictions};
 fn classify_inception() -> anyhow::Result<()> {
     let mut core = Core::new()?;
     let mut model = core.read_model_from_file(
-        &Fixture::graph().to_string_lossy(),
-        &Fixture::weights().to_string_lossy(),
+        &fixture::graph().to_string_lossy(),
+        &fixture::weights().to_string_lossy(),
     )?;
 
     let output_port = model.get_output_by_index(0)?;
     assert_eq!(output_port.get_name()?, "InceptionV3/Predictions/Softmax");
     assert_eq!(model.get_input_by_index(0)?.get_name()?, "input");
 
     // Retrieve the tensor from the test fixtures.
-    let data = fs::read(Fixture::tensor())?;
+    let data = fs::read(fixture::tensor())?;
     let input_shape = Shape::new(&[1, 299, 299, 3])?;
     let element_type = ElementType::F32;
     let mut tensor = Tensor::new(element_type, &input_shape)?;

crates/openvino/tests/classify-mobilenet.rs

@@ -1,10 +1,10 @@
-//! Demonstrates using `openvino-rs` to classify an image using an MobileNet model and a prepared
-//! input tensor. See [README](fixtures/inception/README.md) for details on how this test fixture
-//! was prepared.
+//! Demonstrates using `openvino` to classify an image using an MobileNet model and a prepared input
+//! tensor.
+
 mod fixtures;
 mod util;
 
-use fixtures::mobilenet::Fixture;
+use fixtures::mobilenet as fixture;
 use openvino::{
     prepostprocess, Core, DeviceType, ElementType, Layout, ResizeAlgorithm, Shape, Tensor,
 };
@@ -15,16 +15,16 @@ use util::{Prediction, Predictions};
 fn classify_mobilenet() -> anyhow::Result<()> {
     let mut core = Core::new()?;
     let mut model = core.read_model_from_file(
-        &Fixture::graph().to_string_lossy(),
-        &Fixture::weights().to_string_lossy(),
+        &fixture::graph().to_string_lossy(),
+        &fixture::weights().to_string_lossy(),
     )?;
 
     let output_port = model.get_output_by_index(0)?;
     assert_eq!(output_port.get_name()?, "MobilenetV2/Predictions/Reshape_1");
     assert_eq!(model.get_input_by_index(0)?.get_name()?, "input");
 
     // Retrieve the tensor from the test fixtures.
-    let data = fs::read(Fixture::tensor())?;
+    let data = fs::read(fixture::tensor())?;
     let input_shape = Shape::new(&[1, 224, 224, 3])?;
     let element_type = ElementType::F32;
     let mut tensor = Tensor::new(element_type, &input_shape)?;

crates/openvino/tests/fixtures/alexnet/.gitignore

@@ -0,0 +1,3 @@
+# Ignore all downloaded artifacts in this directory (see `../mod.rs`) except this file.
+*
+!.gitignore