Models update 2 #924

dreadatour · 2025-02-16T06:53:34Z

Another approach to make better Yolo models along with bounding boxes conversions.

Instead of trying to solve unsolvable problem with having one universal model for Bounding Boxes (same for Poses and Segments), it looks like it is better to have dedicated models for every common format, e.g. Yolo. Plus to have really basic and trivial universal models and let users to override them and decide what to do next.

Deploying datachain-documentation with Cloudflare Pages

Latest commit:	`fdc7f1b`
Status:	✅ Deploy successful!
Preview URL:	https://730b1e9f.datachain-documentation.pages.dev
Branch Preview URL:	https://models-update-2.datachain-documentation.pages.dev

View logs

dreadatour · 2025-02-16T06:54:29Z

src/datachain/model/bbox.py

 from pydantic import Field

 from datachain.lib.data_model import DataModel
+from datachain.model.utils import BBoxType, convert_bbox


 class BBox(DataModel):


This model is trivial now. The only thing I've added is type conversion in convert method.

dreadatour · 2025-02-16T06:54:50Z

src/datachain/model/bbox.py

-    The bounding box is defined by two points:
-        - (x1, y1): The top-left corner of the box.
-        - (x2, y2): The bottom-right corner of the box.
+        coords (list[float]): The coordinates of the bounding box.


Floats are more universal here.

why? could you give more details?

Sure! Since it is a generic model, user can decide to store bounding boxes of any format here: COCO, YOLO, Albumentation, etc. Normalized or not. It is just a bare model.

To be honest I don't think we even need these bare models for BBox, Pose, etc, we can leave only models for YOLO for now and add other dedicated models later.

it can break some downstream stuff in data loader / training though? - if it expects ints

also, should we include type into the class? to know is it yolo or something else?

dreadatour · 2025-02-16T06:59:31Z

src/datachain/model/yolo.py

+    name: list[str] = Field(default=[])
+    confidence: list[float] = Field(default=[])
+    box: list[list[float]] = Field(default=[])
+    orig_shape: list[int] = Field(default=[])


I've decided it is also a good idea to keep orig_shape here, to do not read image file when shape is needed.

dreadatour · 2025-02-16T07:00:15Z

src/datachain/model/yolo.py

+            else (0, 0)
+        )
+
+    def to_albumentations(self) -> Iterator[list[float]]:


YoloBox have bounding box type conversion (also below)

dreadatour · 2025-02-16T07:01:16Z

src/datachain/model/yolo.py

+    right_ankle = 16
+
+
+class YoloPose(YoloBox):


Yolo pose result have everything yolo bbox have + poses.

dreadatour · 2025-02-16T07:01:26Z

src/datachain/model/yolo.py

+        )
+
+
+class YoloSeg(YoloBox):


Yolo segment result have everything yolo bbox have + segments.

dreadatour · 2025-02-16T07:01:58Z

src/datachain/model/yolo.py

+        )
+
+
+class YoloCls(DataModel):


Not sure if anyone will use it but Yolo have models so I've added this model here too.

codecov · 2025-02-16T07:13:11Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.19%. Comparing base (c2dd655) to head (fdc7f1b).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #924      +/-   ##
==========================================
+ Coverage   87.66%   88.19%   +0.52%     
==========================================
  Files         130      128       -2     
  Lines       11700    11680      -20     
  Branches     1592     1612      +20     
==========================================
+ Hits        10257    10301      +44     
+ Misses       1043      979      -64     
  Partials      400      400

Flag	Coverage Δ
datachain	`88.11% <100.00%> (+0.52%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

dreadatour · 2025-02-16T07:55:43Z

src/datachain/model/bbox.py

-            title=title,
-            coords=[round(c) for c in coords],
-        )
+    def from_list(coords: Sequence[float], title: str = "") -> "BBox":


Leave this method for now since it is used in many places, but mark it as deprecated. No need in separate method if one can build this model with __init__ method directly.

internally?

I mean instead of using:

BBox.from_list([10, 20, 30, 40])

customers can use:

BBox(coords=[10, 20, 30, 40])

We don't use this model anymore in the datachain source code for now (except for the examples).

shcheklein · 2025-02-18T04:03:06Z

tests/unit/model/test_yolo.py

+    assert_array_almost_equal(yolo_boxes[1], [0.9, 0.9, 0.2, 0.2], decimal=3)
+
+
+def test_yolo_obb():


Q: what to we test with this kind of tests?

We are testing if models were not changed, since other code may rely on them.

shcheklein · 2025-02-18T04:07:06Z

tests/unit/model/test_pose.py

+def test_pose():
+    pose = Pose(x=[10, 20, 30], y=[40, 50, 60])
+    assert pose.model_dump() == {
+        "x": [10.0, 20.0, 30.0],


should these be ints?

Since Pose is now a generic bounding box model I believe we should have floats here. See also my comment here and model declaration here.

shcheklein · 2025-02-18T04:08:41Z

src/datachain/model/bbox.py



 class BBox(DataModel):
    """
    A data model for representing bounding box.

+    Use `datachain.model.Yolo` or for YOLO-specific bounding boxes.


looks like there is a typo

shcheklein · 2025-02-18T04:14:06Z

src/datachain/model/bbox.py

+        Returns:
+            BBox: The bounding box instance.
+        """
+        warn(


we don't need to deprecate anything - we can break it for now

dreadatour added 4 commits February 11, 2025 22:47

Update models

db9edbe

Add validation and tests for the following models: - Bounding box - Pose - Segments Add conversion to/from different bounding box formats, including normalized versions.

Yolo models refactoring

5d0b461

Merge branch 'main' into models-update

515573e

Update examples and docs

375a42b

dreadatour requested review from shcheklein, dmpetrov and a team February 16, 2025 06:53

dreadatour self-assigned this Feb 16, 2025

Add '.DS_Store' to gitignore

a36591a

dreadatour commented Feb 16, 2025

View reviewed changes

dreadatour mentioned this pull request Feb 16, 2025

Update models #919

Closed

1 task

Add 'ultralytics' tests dependency

081e6a1

dreadatour added 2 commits February 16, 2025 14:52

Revert 'BBox.from_list' method

e4ccc3e

Merge branch 'main' into models-update-2

fdc7f1b

dreadatour commented Feb 16, 2025

View reviewed changes

shcheklein reviewed Feb 18, 2025

View reviewed changes

dreadatour mentioned this pull request Feb 26, 2025

Visualize Ultralytics Yolo models #900

Draft

1 task

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Models update 2 #924

Models update 2 #924

dreadatour commented Feb 16, 2025 •

edited

Loading

cloudflare-workers-and-pages bot commented Feb 16, 2025 •

edited

Loading

dreadatour Feb 16, 2025

dreadatour Feb 16, 2025

shcheklein Feb 18, 2025

dreadatour Feb 18, 2025

shcheklein Feb 18, 2025

shcheklein Feb 18, 2025

dreadatour Feb 16, 2025

dreadatour Feb 16, 2025

dreadatour Feb 16, 2025

dreadatour Feb 16, 2025

dreadatour Feb 16, 2025

codecov bot commented Feb 16, 2025 •

edited

Loading

dreadatour Feb 16, 2025

shcheklein Feb 18, 2025

dreadatour Feb 18, 2025

shcheklein Feb 18, 2025

dreadatour Feb 18, 2025

shcheklein Feb 18, 2025

dreadatour Feb 18, 2025

shcheklein Feb 18, 2025

shcheklein Feb 18, 2025

		assert_array_almost_equal(yolo_boxes[1], [0.9, 0.9, 0.2, 0.2], decimal=3)


		def test_yolo_obb():

Models update 2 #924

Are you sure you want to change the base?

Models update 2 #924

Conversation

dreadatour commented Feb 16, 2025 • edited Loading

cloudflare-workers-and-pages bot commented Feb 16, 2025 • edited Loading

Deploying datachain-documentation with Cloudflare Pages

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Feb 16, 2025 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dreadatour commented Feb 16, 2025 •

edited

Loading

cloudflare-workers-and-pages bot commented Feb 16, 2025 •

edited

Loading

codecov bot commented Feb 16, 2025 •

edited

Loading