Feature/yolov26 - #153

Merged

asgersvenning merged 28 commits into

feature/yolov26

Jul 7, 2026

Commits on Jun 6, 2025

Remove some desynced test file pointers
asgersvenning
committed
315a360
Fixed error on saving with SVG due to invalid execution flow logic in TensorPredictions.plot
asgersvenning
committed
b0d6974
Fix --long_format inference
asgersvenning
committed
6f9aee6
Stricter version requirement for ultralytics dependency due to a breaking change in 8.3.125
asgersvenning
committed
ab5361b
Change README install guide from git clone with SSH to HTTPS (re: #121)
asgersvenning
committed
f59cec5

Commits on Jul 9, 2025

Remove tile and image tensors from intermediate `Results` objects to avoid memory-leak and OOM on large images
asgersvenning
committed
72f407a

Commits on Oct 6, 2025

Merge branch 'develop' of github.com:darsa-group/flat-bug into develop
qgeissmann
committed
1addfac

Commits on Oct 8, 2025

work on #134
qgeissmann
committed
8e23f7f

Commits on Oct 12, 2025

improved
qgeissmann
committed
22acf67
Merge branch 'feature_mask_mask_refiner' of github.com:darsa-group/flat-bug into develop
qgeissmann
committed
135eb6d

Commits on Jan 29, 2026

Merge branch 'develop' of github.com:darsa-group/flat-bug into develop
qgeissmann
committed
18f4044

Commits on Feb 5, 2026

Merge branch 'develop' of github.com:darsa-group/flat-bug into develop
qgeissmann
committed
1c26793

Commits on May 7, 2026

Merge branch 'develop' of github.com:darsa-group/flat-bug into develop
qgeissmann
committed
24f46e2

Commits on May 13, 2026

Merge branch 'develop' of github.com:darsa-group/flat-bug into develop
qgeissmann
committed
7605ab5
Add YOLOv26 support for training and inference

qgeissmann
and
claude
committed
e5d4da1
Simplify Predictor model loading, fix ultralytics 8.4.x compatibility

qgeissmann
and
claude
committed
a2e7c6d

Commits on May 14, 2026

Name training runs as fb_{size}_{timestamp} and update N40S config

qgeissmann
and
claude
committed
75c1305

Commits on May 18, 2026

Default to YOLOv26 for training, fix model names in configs

qgeissmann
and
claude
committed
75e0d97
Download pretrained weights explicitly when not present locally

qgeissmann
and
claude
committed
3a818f6

Commits on May 19, 2026

Free GPU cache before validation to avoid OOM from fragmentation

qgeissmann
and
claude
committed
e551822
Cap validation batch size at training batch size to avoid OOM

qgeissmann
and
claude
committed
b00fba5

Commits on May 20, 2026

Use batch size 1 for validation to avoid OOM

qgeissmann
and
claude
committed
8a19b4b
Use PNG compress_level=1 for crop saving to reduce write time

qgeissmann
and
claude
committed
87e5692

Commits on Jun 4, 2026

ibid
qgeissmann
committed
463d7c6

Commits on Jun 25, 2026

Suppress validation loss to fix OOM with YOLOv26 one2many head

qgeissmann
and
claude
committed
9570493

Commits on Jun 26, 2026

Fix tensor shape mismatch in validation loss no-op

qgeissmann
and
claude
committed
289f8b3

Commits on Jul 7, 2026