Repository navigation
YOLOVAL: preparing game-model predictions as CVAT XML for human review (paid ONNX models) #11435
bbshdh56
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
I publish YOLOVAL, game-specific object-detection models in ONNX (.onnx) format. The model packages are paid. I'm sharing a small CVAT-format preparation workflow and the public visual portfolio.
I took saved predictions from 14 full screenshots of Delta Force and Arena Breakout and used Datumaro 1.13.12 to export CVAT XML 1.1, then re-imported the result locally.
One useful detail: the CVAT exporter skips undeclared custom attributes by default. Declaring score, annotation_source and review_status on each label category preserved the metadata:
The local round trip retained 14 images, 15 predicted boxes, and 5 images without boxes. Class names, confidence scores and the unreviewed marker survived; the maximum bounding-box coordinate difference was 0 pixels. Game-prefixed labels kept the two model vocabularies separate.
The image copies and XML stayed local. Validation covered XML export and re-import with Datumaro; a running CVAT server upload was outside this check. These are unreviewed model predictions, intended as a starting point for human annotation review, not ground-truth labels or an accuracy benchmark.
The broader English model portfolio contains 32 detection examples across 15 game/category groups, retaining the .onnx filenames and colorful yoloval.cc.cd watermarks. It is separate from the 14-image file-format check above.
Model catalog: https://yoloval.cc.cd/
Patreon shop — model details, prices and downloads: https://www.patreon.com/c/sugaraigogo/shop
Public visual showcase: https://github.com/bbshdh56/yoloval-onnx-models
All reactions