The usage of the pre-trained model

Your work is of great value. I have two issues to consult regarding its usage.
1. During runtime, in the created virtual environment, `hydra` can be imported, and it is confirmed that `hydra-core-1.3.2` already exists. However, an error of `ModuleNotFoundError: No module named 'hydra'` is still reported during runtime. How should this be resolved?
2. How can the `pretrained image_encoder` of SAM 2.1 be used alone to extract the features of the source image? What should the input be and what output will be obtained? Thank you for your explanation.
3.

Reproduce the hydra import failure in the created virtual environment, then inspect the repository's example notebooks for SAM 2.1 pretrained-model usage. Document the environment resolution and specify the pretrained image_encoder input and returned output, with enough detail for users to follow the workflow.

Written by the indexing model from the issue text.