[v1.x] Refactor onnx tests for object classification, add object detection tests #19802
Conversation
…plits attribute for onnx Split operator. Add operator unit test for SliceChannel.
…object detection tests with currently supported models.
|
Hey @josephevans , Thanks for submitting the PR
CI supported jobs: [windows-cpu, centos-cpu, clang, unix-cpu, website, unix-gpu, edge, centos-gpu, windows-gpu, sanity, miscellaneous] Note: |
lanking520
added
pr-awaiting-testing
lanking520
added
pr-awaiting-testing
lanking520
added
pr-awaiting-testing
| res = softmax(np.array(raw_result)).tolist() | ||
| class_idx = np.argmax(res) | ||
| assert(class_idx in accepted_ids) | ||
| test_image_urls = ['https://raw.githubusercontent.com/zhreshold/mxnet-ssd/master/data/demo/dog.jpg'] |
There was a problem hiding this comment.
shall we consider adding more test images?
There was a problem hiding this comment.
Yeah, we will soon in upcoming PRs.
There was a problem hiding this comment.
Also we should have tests which avoid depending on extrenal sources
There was a problem hiding this comment.
In upcoming PRs for these onnxruntime tests, let's migrate to use either an s3 bucket for test data or check them into the repo. the first is ideal.
There was a problem hiding this comment.
We can also create a second repo just for resources. I would actually prefer that because this would allow everyone to fully replicate the setup even it the s3 bucket goes away
lanking520
added
pr-awaiting-review
lanking520
added
pr-awaiting-testing
|
Let's move the test images to a separate repo/s3 in the next pr. Merging this for now, thanks @josephevans |
Description
Refactor onnx tests for object classification, add object detection tests.
Also fix a few ONNX export operators: