You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
YOLOVAL: a two-image sliced-inference check with game-specific ONNX models (paid)
#1508
I publish YOLOVAL, game-specific detection models in ONNX (.onnx) format. The model packages are paid. Here is a small SAHI test with two exported models, followed by the public visual portfolio.
I ran SAHI 0.14.0, Ultralytics 8.3.179, and ONNX Runtime 1.24.3 on CPU, using one source screenshot per model.
Model
Input / tile size
Tiles at 20% overlap
Full-image boxes
Sliced-only boxes after merging
Delta Force
640 x 640
12
1
1
Arena Breakout
320 x 320
40
3
0
Settings: confidence 0.25, batch size 1, standard/full-image prediction disabled for the sliced run, class-aware NMS merging with IoU 0.45. The Ultralytics adapter used task="detect" and the model's input size.
For these two examples, slicing did not increase the detection count. This is a useful reminder to inspect full-image and sliced results separately for a domain-specific detector. There are no ground-truth annotations in this small check, so the counts do not establish precision, recall, or a performance improvement. Both original ONNX file hashes remained unchanged.
The English overview below is a separate portfolio containing 32 detection examples across 15 game/category groups, with .onnx filenames and colorful yoloval.cc.cd watermarks.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
I publish YOLOVAL, game-specific detection models in ONNX (.onnx) format. The model packages are paid. Here is a small SAHI test with two exported models, followed by the public visual portfolio.
I ran SAHI 0.14.0, Ultralytics 8.3.179, and ONNX Runtime 1.24.3 on CPU, using one source screenshot per model.
Settings: confidence 0.25, batch size 1, standard/full-image prediction disabled for the sliced run, class-aware NMS merging with IoU 0.45. The Ultralytics adapter used task="detect" and the model's input size.
The relevant slicing call was:
For these two examples, slicing did not increase the detection count. This is a useful reminder to inspect full-image and sliced results separately for a domain-specific detector. There are no ground-truth annotations in this small check, so the counts do not establish precision, recall, or a performance improvement. Both original ONNX file hashes remained unchanged.
The English overview below is a separate portfolio containing 32 detection examples across 15 game/category groups, with .onnx filenames and colorful yoloval.cc.cd watermarks.
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