DepthAI 3.8.0: getModelFromZoo() fails in official examples despite working Internet connection

[Wangzr1222]

Environment

• OS: Windows 10 x64
• Python: 3.10.20 (Anaconda)
• depthai: 3.8.0
• depthai-nodes: latest
• oak-examples: latest main
• Commit: f07aa4043ac188509ee06844a943bf6b2dd8fbc2
• Device: OAK-D (Platform.RVC2)

Problem

The official examples fail when calling dai.getModelFromZoo().

I tested the following official examples:

• apps/default-app
• neural-networks/generic-example

Both fail at exactly the same line:

nn_archive = dai.NNArchive(dai.getModelFromZoo(model_description))

with the following error:

RuntimeError: Model not available. No internet connection available.
Could not connect to host:
https://easyml.cloud.luxonis.com/models/api/v1/health/

What I have verified

I have checked the following:

• Internet connection is working normally.
• The OAK device is detected correctly.
• The YAML model description is loaded successfully.
• curl can access the health endpoint.
• Python requests can also access the same endpoint.
• OpenSSL and certifi are installed and working correctly.
• No HTTP/HTTPS proxy is configured.
• The same error occurs in multiple official examples, so it does not appear to be application-specific.

curl

curl https://easyml.cloud.luxonis.com/models/api/v1/health/

returns

{"version":"development"}

Python requests

import requests

r = requests.get("https://easyml.cloud.luxonis.com/models/api/v1/health/")
print(r.status_code)
print(r.text)

returns

200
{"version":"development"}

YAML loading

The YAML file loads successfully:

model_description = dai.NNModelDescription.fromYamlFile(
    "./depthai_models/yolov6_nano_r2_coco.RVC2.yaml"
)

print(model_description)

Output:

NNModelDescription [
  model: luxonis/yolov6-nano:r2-coco-512x288
  platform: RVC2
]

The failure only happens when calling:

dai.getModelFromZoo(model_description)

Additional observation

I also tested the API manually.

The health endpoint works:

curl https://easyml.cloud.luxonis.com/models/api/v1/health/

However,

curl "https://easyml.cloud.luxonis.com/models/api/v1/models/luxonis/yolov6-nano:r2-coco-512x288"

returns

{"detail":"Not Found"}

Could this indicate that the model name or the Model Zoo API has changed?

Expected behavior

The official examples should automatically download the required model and run successfully.

Actual behavior

dai.getModelFromZoo() always throws

Model not available.
No internet connection available.

even though the machine has working Internet access.

Could you please help determine whether this is a Model Zoo API issue or a bug in getModelFromZoo()?

Thank you!

[klemen1999]

Hi @Wangzr1222,

I’m not able to reproduce the issue on my side. Could you try running the examples again? It’s possible the timing was unlucky and you encountered a brief period of downtime in our Model Zoo.

[Wangzr1222]

Hi, thanks for your reply.

I tried running the examples again today, but unfortunately I still get the same error.

The OAK device is detected correctly now, and the pipeline starts to be created successfully. However, it always fails when calling dai.getModelFromZoo().

I tested both official examples:

• apps/default-app
• neural-networks/generic-example

Both fail at exactly the same line:

nn_archive = dai.NNArchive(dai.getModelFromZoo(model_description))

with the following error:

Creating pipeline...
Traceback (most recent call last):
  File ".../main.py", line 43, in <module>
    nn_archive = dai.NNArchive(dai.getModelFromZoo(model_description))
RuntimeError: Model not available. No internet connection available.
Could not connect to host: https://easyml.cloud.luxonis.com/models/api/v1/health/

I have verified the following:

• The machine has a working Internet connection.
• curl https://easyml.cloud.luxonis.com/models/api/v1/health/ returns:

{"version":"development"}

• Python requests can access the same endpoint successfully (HTTP 200).
• The OAK device is detected correctly (Platform.RVC2).
• I am using:
  • depthai==3.8.0
  • the latest oak-examples (commit f07aa4043ac188509ee06844a943bf6b2dd8fbc2)

Since multiple official examples fail at the same getModelFromZoo() call, it does not seem to be specific to a single application.

Is there any way to enable verbose/debug logging for getModelFromZoo() or the Model Zoo downloader? I'd be happy to provide additional logs if that helps identify the issue.

Thank you!

[klemen1999]

@Wangzr1222 Currently we don't see any downtime on our Model Zoo backend. But what could be the problem is that the check if Zoo is available performed by DepthAI takes more than 1s for you. This timeout can be configured through the DEPTHAI_ZOO_INTERNET_CHECK_TIMEOUT environment variable.
Could you please try using higher timeout e.g. export DEPTHAI_ZOO_INTERNET_CHECK_TIMEOUT=5000 and check if your issue is stil present afterwards?

Klemen

[klemen1999]

Hi @Wangzr1222
I just wanted to check if the suggested fix helped you and if I can close the issue?

Thanks,
Klemen