Add object detection to inference client

[dulayjm]

Add object-detection to HuggingFace🤗 Hub

References Issue #1539

This is an ongoing list of model tasks to implement in the Hugging Face Hub inference client. Each task is planned to be its own PR. The task for this is object-detection.

** Key Features **

• Modifies _client.py to call an object detection model. The example model is DETR ResNet-50.
• Modifies testing and documentation to reflect changes.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[Wauplin]

Very clean implementation! Thanks for working on it @dulayjm !

I tested it locally and re-recorded the test (there were 3 requests made, making the file quite huge). Worked like a charm for me at first use :D If everything's fine in the CI, it is ready to be merged straight away 🤗

[dulayjm]

@Wauplin Thanks!

The code seems to still fail the CI, but I have a question about that. On the Python Quality check, it fails with some messages at the end. In particular, the LocalTokenNotFoundError is one I'm seeing a couple of times. Is there something that needs to be different about the VCR testing?

=========================== short test summary info ============================
FAILED ../tests/test_commit_scheduler.py::TestCommitScheduler::test_mocked_push_to_hub - AssertionError: 2 not greater than 2
FAILED ../tests/test_hf_api.py::HfApiListFilesInfoTest::test_list_files_with_expand - AssertionError: False is not true
FAILED ../tests/test_webhooks_server.py::TestWebhooksServerRun::test_run_parse_payload - AssertionError: 422 != 200
ERROR ../tests/test_hf_api.py::TestSpaceAPIProduction::test_manage_secrets - huggingface_hub.utils._headers.LocalTokenNotFoundError: Token is required (`token=True`), but no token found. You need to provide a token or be logged in to Hugging Face with `huggingface-cli login` or `huggingface_hub.login`. See https://huggingface.co/settings/tokens.
ERROR ../tests/test_hf_api.py::TestSpaceAPIProduction::test_pause_and_restart_space - huggingface_hub.utils._headers.LocalTokenNotFoundError: Token is required (`token=True`), but no token found. You need to provide a token or be logged in to Hugging Face with `huggingface-cli login` or `huggingface_hub.login`. See https://huggingface.co/settings/tokens.
ERROR ../tests/test_hf_api.py::TestSpaceAPIProduction::test_space_runtime - huggingface_hub.utils._headers.LocalTokenNotFoundError: Token is required (`token=True`), but no token found. You need to provide a token or be logged in to Hugging Face with `huggingface-cli login` or `huggingface_hub.login`. See https://huggingface.co/settings/tokens.
= 3 failed, 660 passed, 40 skipped, 60 deselected, 120 warnings, 3 errors in 342.14s (0:05:42) =
Error: Process completed with exit code 1.

[Wauplin]

@dulayjm All failing tests are unrelated. Some tests actually require a production token that is set as a secret on the repo. Since your PR comes from a fork -and to avoid any leaks-, Github doesn't seem to be forwarding the secret values. I thought that when a maintainer (me) triggers it manually the secrets were passed but that doesn't seem to be the case. I'll investigate to check what I could do in PRs from forks (most likely skip the related tests).

But anyway, PR is good and related test worked so let's merge it! :)

[dulayjm]

Awesome, thank you so much!

[julien-c]

This is really cool! Random unsollicited question, did you try out the huggingface.js implementation? https://huggingface.co/docs/huggingface.js/inference/classes/HfInference#objectdetection

(same client, but in JS)

[dulayjm]

Thank you!!

No, I haven't checked out the JS library, but I'll give it a look! 🤗