Skip to content

Optimize Deployment and Extraction #60

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
RyanDoesMath merged 42 commits into from
Mar 7, 2025
Merged

Optimize Deployment and Extraction #60

RyanDoesMath merged 42 commits into from
Mar 7, 2025

Conversation

@RyanDoesMath
Copy link
Member

@RyanDoesMath RyanDoesMath commented Mar 7, 2025

This pull request makes a few changes.

  1. Onnx models are now the default for running the program. Ultralytics will become a dev dependency that will not be packaged into the final product.
  2. ObjectDetectionModel is now a protocol instead of an abstractbaseclass.

This will (hopefully) make our deployments much smaller.

RyanDoesMath and others added 30 commits February 28, 2025 17:43
@RyanDoesMath
Create onnx_yolov11_detection.py.
214e9b1
@RyanDoesMath
Import object detection model.
df03945 
Didn't do much because I am switching computers after this commit.
Add init and sigmoid fns.
bcd7fbb
add method to preprocess images.
5a652d0
Add constants.
4dc2900
Stub out the __call__ method.
d14362f
Implement __call__ and stub out the detect fn.
5ce72f3
Add postprocess_results fn and fill out detect fn.
6818089
reorder methods.
0e0e1ff
Add postprocessing method.
fce1164 
This function is impenetrable... Need to refactor soon.
Import nms.
871b065
@RyanDoesMath
Write docstring for postprocess_results.
2ebdabb
@RyanDoesMath
Stub out fn for finding class names.
ede2e02
@RyanDoesMath
Make enough changes to make the file run.
220c5e3 
The code is still slow compared to what it should be (runs in 2 seconds, but the actual model runs in about 0.4 seconds). This is due to postprocessing being slow in python. One last thing to try is ultralytics yolov11, but loaded from an onnx model.
@RyanDoesMath
Change pyproject.toml
3f6fc3a
@RyanDoesMath
Switch to load models once.
5b2ab39
@RyanDoesMath
Majorly revise OnnxYolov11Detection to improve efficiency.
c36029d
@RyanDoesMath
Add dockerfile.
01242b0
@RyanDoesMath
Merge pull request #57 from Paper-Chart-Extraction-Project/add-docker
608b5ec 
Add dockerfile.
@RyanDoesMath
Write docstring for the load_classes static method.
da91391
@RyanDoesMath
Separate reading the yaml file from reading the config.
a2ee2c8
@RyanDoesMath
Implement the load_classes static method.
290cbd6
@RyanDoesMath
Change output datatype of load_classes.
88c03a6
@RyanDoesMath
Add conversion from numpy ndarray to list of Detections in detect().
0e4c92b
@RyanDoesMath
Add type and docstring entry for err_msg parameter in read_yaml_file.
f73a739
@RyanDoesMath
Remove unnecessary sigmoid fn call.
49b1737
@RyanDoesMath
Switch draw_detections to work with Detections and not numpy ndarrays.
95dade6
@RyanDoesMath
Create onnx_yolov11_pose_single.py
5f11b93
@RyanDoesMath
Remove unused consts.
6ac8e7f
@RyanDoesMath
copy over all the methods from OnnxYolov11Detection.
38d8327
RyanDoesMath and others added 11 commits March 6, 2025 13:12
@RyanDoesMath
Convert output of onnx keypoint model to Detections.
89a0be7 
Also, filtered out all detections whose keypoint is not in the bounding box. This I learned is actually a feature of these pose models, not a bug.
@RyanDoesMath
Add optional parameter resize_method to preprocess_image.
5bb08e5 
This parameter lets the user determine if they want resizing or letterboxing.
@RyanDoesMath
Merge pull request #58 from Paper-Chart-Extraction-Project/add-onnx-m…
c91e85a 
…odels

Add onnx models
@RyanDoesMath
Change ObjectDetectionModel from an interface to a protocol.
60e9367 
ObjectDetectionModel was acting a bit too much like a parent class, so its two constructor methods were removed and the class was changed to a protocol which enforces that an implementer's object can be called to produce a list of Detection objects.
@RyanDoesMath
Remove type hint for PIL image in ObjectDetectionModel protocol.
91e00a2 
This is in preparation for the switch to numpy arrays.
@RyanDoesMath
Remove unused constructors from Onnx model classes.
6c98113 
These existed to satisfy the interface, but weren't implemented.
@RyanDoesMath
Make various small bugfixes.
e409949
@RyanDoesMath
Merge pull request #59 from Paper-Chart-Extraction-Project/switch-obj…
b634c2d 
…-detection-model-to-protocol

Switch ObjectDetectionModel to protocol from an abstractbaseclass
@RyanDoesMath
Fix very obvious error in read_config.
4694a21
@RyanDoesMath
Make a myriad of changes to make the program run.
1f8ff20 
Errors persist with digitization.
@RyanDoesMath
Add fn to rescale detections to original image size.
f218e33
@RyanDoesMath RyanDoesMath self-assigned this Mar 7, 2025
@RyanDoesMath RyanDoesMath merged commit c925b07 into main Mar 7, 2025
@RyanDoesMath RyanDoesMath deleted the optimize-extraction branch March 7, 2025 11:39
This was linked to issues Mar 7, 2025
Add an ObjectDetection subclass for ONNX models. #54
Closed
Refactor ObjectDetectionModel into a Protocol #55
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

@RyanDoesMath RyanDoesMath

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Refactor ObjectDetectionModel into a Protocol Add an ObjectDetection subclass for ONNX models.

2 participants

@RyanDoesMath