Pandede

You may apply knowledge distillation, suggest distill the knowledge from FaceNet to MobileNet

Regardless of the performance and time-consuming, training a NN-based classifier is better than SVM classifier. Training a SVM multiclass classifier take large amount of time because it's breaking down to...

You may see the list of available models in `__model_factory` in `torchreid/models/__init__.py`, therefore the answer is YES: `osnet_x1_0` is available.

Use optional argument `duration` ([Reference](https://imageio.readthedocs.io/en/v2.4.1/format_gif-pil.html)), it accepts `float` for all frames or `list` for one value of each frame. Pausing the GIF between loops can be done via this tricky...

Any minimum code which can reproduce this error?

Use the marker `needs_internet` to handle the internet interactions of `test_images` is a perfect approach. @FirefoxMetzger How do you think about it?

> I think this is a useful feature to have. I'll try to include it in v4.0 if possible. If it's too much work, then I'll target 4.1. Is that...

~~The plugin `ffmpeg` is implemented based on `imageio-ffmpeg`.~~ ~~You may transfer the issue to [this repository](https://github.com/imageio/imageio-ffmpeg).~~ **Update:** @jsakkos-curio You may pass `ffpmeg_params=['-loglevel', 'error']` towards `imread` to bypass the warning message...

`numpy` v1 and v2 is incompatible, there is many [breaking changes](https://numpy.org/devdocs/numpy_2_0_migration_guide.html#changes-to-namespaces) in API usage and namespace, that's why the tests broke.

May you provide some examples for your ideas?