Depth estimation workflow block #1175
Conversation
reiffd7
requested review from
capjamesg,
PawelPeczek-Roboflow,
grzegorz-roboflow,
hansent,
yeldarby and
probicheaux
as code owners
…/inference into depth-estimation-workflow-block
…boflow. replaced model id with new nomenclature: depth-anything-v2/small
| "3. Setting the token in your environment: export HUGGING_FACE_HUB_TOKEN=your_token_here\n" | ||
| "Or by logging in with: huggingface-cli login" | ||
| ) from e | ||
| print(f"Error initializing depth estimation model: {str(e)}") |
There was a problem hiding this comment.
please use logger instead
|
|
||
| def __init__( | ||
| self, model_id, *args, dtype=None, huggingface_token=HUGGINGFACE_TOKEN, **kwargs | ||
| self, *args, dtype=None, huggingface_token=HUGGINGFACE_TOKEN, **kwargs |
There was a problem hiding this comment.
please do not remove this parameter without agreement from the code owner (probably @probicheaux would be able to judge best).
| self.initialize_model() | ||
|
|
||
| # Try to initialize model from cache first | ||
| try: |
There was a problem hiding this comment.
please do not change base class as that may imply changes to many other models
|
@PawelPeczek-Roboflow I got rid of structural changes in transformers base class and am passing integration tests. |
| if model_type not in self.registry_dict: | ||
| raise ModelNotRecognisedError(f"Model type not supported: {model_type}") | ||
| return self.registry_dict[model_type] | ||
| try: |
There was a problem hiding this comment.
am I right that this change does nothing compared to what it was?
There was a problem hiding this comment.
LGTM, apart from one not needed change: https://github.com/roboflow/inference/pull/1175/files#r2047394190
There was a problem hiding this comment.
LGTM, but conflicts needs to be resolved
@grzegorz-roboflow / @hansent - since I am done for today, would u approve when @reiffd7 resolves conflicts?
…ation-workflow-block
…h-estimation-workflow-block`) In order to optimize the performance of the `add_model` and `get_model` methods, we need to minimize the time spent on certain operations and reduce the number of loggings wherever possible. **Changes made:** 1. In the `get_model` method of `ModelRegistry`, a single `try/except` block replaces the if-check and separate raise to catch `KeyError`, thereby combining lookup and retrieval into a single step. This optimizes the process of fetching the model class and raising the error when necessary. 2. In the `add_model` method of `ModelManager`, logging messages are minimized and combined to reduce the overhead caused by frequent logging. The change reduces unnecessary calls to `logger.debug()`. **Line profiling improvements:** - For `get_model`: Combined lookup and retrieval operations reduce steps. - For `add_model`: Consolidating `logger.debug()` calls minimizes overhead and redundant operations.
⚡️ Codeflash found optimizations for this PR📄 11% (0.11x) speedup for
|
Description
Implemented a new Depth Estimation workflow block that leverages Depth Anything V2 model to perform depth estimation on images. This block provides a powerful tool for analyzing spatial relationships and creating depth maps from 2D images.
The implementation includes:
The block is implemented in
inference/core/workflows/core_steps/models/foundation/depth_estimation/v1.py.Type of change
How has this change been tested, please provide a testcase or example of how you tested the change?
The implementation was tested through:
Any specific deployment considerations
Docs