`ultralytics 8.2.10` add Classify and OBB Tasks to `Results.summary()`

glenn-jocher commented 1 week ago

Found two bugs in OBB results and Classify dataloading in this PR:

Results.summary() lacks OBB handling @Laughing-q
Classify models are unable to accept numpy or list inputs it seems EDIT: Fixed, turns out the problem was when passing float numpy arrays (must be UINT8)

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancements in testing flexibility and coverage for Ultralytics models 🚀

📊 Key Changes

Introduced MODELS constant: Streamlines model management by centralizing model names in a single location.
Parameterized testing: Simplified test_predict_img and test_results to iterate over all models, improving test coverage and maintainability.
Code cleanup: Removed redundant code and streamlined testing functions, making the codebase cleaner and easier to understand.

🎯 Purpose & Impact

For Developers: Simplifies adding new models and writing tests, leading to faster development and better stability.
For Users: Ensures a more robust and reliable framework by enhancing test coverage, indirectly improving user experience through higher quality assurance.
Potential Impact: These changes could lead to quicker iterations for new features and models, and better performance and stability across the board.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved testing and model result formats in Ultralytics software.

📊 Key Changes

Added parameterization in tests for image predictions and result formats to expand testing across multiple models.
Introduced MODELS set for dynamic testing based on available models.
Enhanced result summarization logic for better clarity and added support for object bounding boxes (OBB) and classification probabilities.
Minor version bump from 8.2.9 to 8.2.10.

🎯 Purpose & Impact

Broader Coverage: Parameterized tests ensure a wide range of models are automatically verified, enhancing the robustness of the software.
Flexibility and Ease: By dynamically determining models for tests, developers are saved from manual updates, making the codebase more maintainable.
Clearer Outputs: The updated result formats provide users with clearer insights into model predictions, potentially improving interpretability for end applications.
Version Increment: Indicates ongoing progress and improvements, signaling to users that they are getting the latest features and fixes.

These changes aim to streamline testing processes, improve code maintainability, and enhance the user experience by providing clearer insights into model outputs. 🚀

codecov[bot] commented 1 week ago

Codecov Report

Attention: Patch coverage is 11.76471% with 15 lines in your changes are missing coverage. Please review.

Project coverage is 37.38%. Comparing base (f2f1afd) to head (be722fe).

Files	Patch %	Lines
ultralytics/engine/results.py	0.00%	15 Missing :warning:

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #11653 +/- ## ======================================= Coverage 37.37% 37.38% ======================================= Files 122 122 Lines 15584 15590 +6 ======================================= + Hits 5825 5828 +3 - Misses 9759 9762 +3 ``` | [Flag](https://app.codecov.io/gh/ultralytics/ultralytics/pull/11653/flags?src=pr&el=flags&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ultralytics) | Coverage Δ | | |---|---|---| | [GPU](https://app.codecov.io/gh/ultralytics/ultralytics/pull/11653/flags?src=pr&el=flag&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ultralytics) | `37.38% <11.76%> (+<0.01%)` | :arrow_up: | Flags with carried forward coverage won't be shown. [Click here](https://docs.codecov.io/docs/carryforward-flags?utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=ultralytics#carryforward-flags-in-the-pull-request-comment) to find out more.

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

glenn-jocher commented 1 week ago

@Laughing-q not sure if OBB Results should include an 'xyxy' box? We should remove this attribute no?

Ah nevermind, I see this, I understand now, all good!

glenn-jocher commented 1 week ago

@Laughing-q could you try to update Results.summary() to enable OBB and Classify model summaries? The summaries are essentially the default JSON return for our inference API. They work correctly now for Detect, Segment and Pose models but need additional handling for Classify and OBB in order for the newly parameterized test_results test to pass.

Thank you!

Laughing-q commented 1 week ago

@glenn-jocher Done in https://github.com/ultralytics/ultralytics/pull/11653/commits/87f46e77128af9125b9da30033ed3c8ad56d3f1e. Currently classify model would only output names, confidence, class keys(no box), and obb by default returns xyxyxyxy format for box with keys x1, y1, x2, y2, x3, y3, x4, y4. local test script:

from ultralytics import YOLO
model = YOLO("yolov8n-cls.pt")
results = model.predict(source="ultralytics/assets/bus.jpg")
summary = results[0].summary()[0]
print(summary)

model = YOLO("yolov8n.pt")
results = model.predict(source="ultralytics/assets/bus.jpg")
summary = results[0].summary()[0]
print(summary)

model = YOLO("yolov8n-obb.pt")
results = model.predict(source="../datasets/dota8/images/train/P0861__1024__0___1648.jpg")

summary = results[0].summary()[0]
print(summary)

output:

image 1/1 /home/laughing/codes/ultralytics/ultralytics/assets/bus.jpg: 224x224 minibus 0.47, police_van 0.25, ambulance 0.05, recreational_vehicle 0.04, bullet_trai
n 0.02, 1.4ms
Speed: 0.8ms preprocess, 1.4ms inference, 0.0ms postprocess per image at shape (1, 3, 224, 224)
{'name': 'minibus', 'class': 654, 'confidence': 0.46673}

image 1/1 /home/laughing/codes/ultralytics/ultralytics/assets/bus.jpg: 640x480 4 persons, 1 bus, 1 stop sign, 47.7ms
Speed: 1.2ms preprocess, 47.7ms inference, 1.4ms postprocess per image at shape (1, 3, 640, 480)
{'name': 'bus', 'class': 5, 'confidence': 0.87327, 'box': {'x1': 22.84789, 'y1': 231.27481, 'x2': 804.99994, 'y2': 756.82684}}

image 1/1 /home/laughing/codes/datasets/dota8/images/train/P0861__1024__0___1648.jpg: 1024x1024 6.7ms
Speed: 2.2ms preprocess, 6.7ms inference, 3.1ms postprocess per image at shape (1, 3, 1024, 1024)
{'name': 'harbor', 'class': 7, 'confidence': 0.87786, 'box': {'x1': 326.19873, 'y1': 202.04541, 'x2': 441.58606, 'y2': 232.54843, 'x3': 500.82678, 'y3': 8.4515, 'x4
': 385.43945, 'y4': -22.05151}}

EDIT: forgot about the normalize option, fixed in https://github.com/ultralytics/ultralytics/pull/11653/commits/b3070d5540d41f39bc710fc0ea4e3650ad197c06.

glenn-jocher commented 1 week ago

@Laughing-q nice work!! PR merged as ultralytics 8.2.10 :)

ultralytics / ultralytics