From f65e06bea913aa185c89255cdba5bec6f4d10bad Mon Sep 17 00:00:00 2001 From: Murat Raimbekov Date: Tue, 10 Feb 2026 13:55:26 +0600 Subject: [PATCH] Add footnote clarifying param/FLOP counts after `model.fuse()` (#23601) Co-authored-by: Glenn Jocher Co-authored-by: UltralyticsAssistant Co-authored-by: Jing Qiu <61612323+Laughing-q@users.noreply.github.com> --- docs/en/models/yolo26.md | 2 ++ docs/en/models/yolov10.md | 2 ++ docs/en/tasks/classify.md | 1 + docs/en/tasks/detect.md | 1 + docs/en/tasks/obb.md | 1 + docs/en/tasks/pose.md | 1 + docs/en/tasks/segment.md | 1 + 7 files changed, 9 insertions(+) diff --git a/docs/en/models/yolo26.md b/docs/en/models/yolo26.md index f280565d..6b3a070f 100644 --- a/docs/en/models/yolo26.md +++ b/docs/en/models/yolo26.md @@ -105,6 +105,8 @@ This unified framework ensures YOLO26 is applicable across real-time detection, --8<-- "docs/macros/yolo-obb-perf.md" +_Params and FLOPs values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers and removes the auxiliary one-to-many detection head. Pretrained checkpoints retain the full training architecture and may show higher counts._ + --- ## Usage Examples diff --git a/docs/en/models/yolov10.md b/docs/en/models/yolov10.md index 34e7c3c8..bbe09c33 100644 --- a/docs/en/models/yolov10.md +++ b/docs/en/models/yolov10.md @@ -149,6 +149,8 @@ Compared to other state-of-the-art detectors: [5]: https://github.com/ultralytics/assets/releases/download/v8.4.0/yolov10l.pt [6]: https://github.com/ultralytics/assets/releases/download/v8.4.0/yolov10x.pt +_Params and FLOPs values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers and removes the auxiliary one-to-many detection head. Pretrained checkpoints retain the full training architecture and may show higher counts._ + ## Usage Examples For predicting new images with YOLOv10: diff --git a/docs/en/tasks/classify.md b/docs/en/tasks/classify.md index 33e6ae62..d10a09f8 100644 --- a/docs/en/tasks/classify.md +++ b/docs/en/tasks/classify.md @@ -38,6 +38,7 @@ YOLO26 pretrained Classify models are shown here. Detect, Segment, and Pose mode - **acc** values are model accuracies on the [ImageNet](https://www.image-net.org/) dataset validation set.
Reproduce by `yolo val classify data=path/to/ImageNet device=0` - **Speed** averaged over ImageNet val images using an [Amazon EC2 P4d](https://aws.amazon.com/ec2/instance-types/p4/) instance.
Reproduce by `yolo val classify data=path/to/ImageNet batch=1 device=0|cpu` +- **Params** and **FLOPs** values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers. Pretrained checkpoints retain the full training architecture and may show higher counts. ## Train diff --git a/docs/en/tasks/detect.md b/docs/en/tasks/detect.md index fae76c34..17ebd099 100644 --- a/docs/en/tasks/detect.md +++ b/docs/en/tasks/detect.md @@ -37,6 +37,7 @@ YOLO26 pretrained Detect models are shown here. Detect, Segment, and Pose models - **mAPval** values are for single-model single-scale on [COCO val2017](https://cocodataset.org/) dataset.
Reproduce by `yolo val detect data=coco.yaml device=0` - **Speed** averaged over COCO val images using an [Amazon EC2 P4d](https://aws.amazon.com/ec2/instance-types/p4/) instance.
Reproduce by `yolo val detect data=coco.yaml batch=1 device=0|cpu` +- **Params** and **FLOPs** values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers and, for end2end models, removes the auxiliary one-to-many detection head. Pretrained checkpoints retain the full training architecture and may show higher counts. ## Train diff --git a/docs/en/tasks/obb.md b/docs/en/tasks/obb.md index 11eb7f91..0eb43049 100644 --- a/docs/en/tasks/obb.md +++ b/docs/en/tasks/obb.md @@ -46,6 +46,7 @@ YOLO26 pretrained OBB models are shown here, which are pretrained on the [DOTAv1 - **mAPtest** values are for single-model multiscale on [DOTAv1](https://captain-whu.github.io/DOTA/index.html) dataset.
Reproduce by `yolo val obb data=DOTAv1.yaml device=0 split=test` and submit merged results to [DOTA evaluation](https://captain-whu.github.io/DOTA/evaluation.html). - **Speed** averaged over DOTAv1 val images using an [Amazon EC2 P4d](https://aws.amazon.com/ec2/instance-types/p4/) instance.
Reproduce by `yolo val obb data=DOTAv1.yaml batch=1 device=0|cpu` +- **Params** and **FLOPs** values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers and, for end2end models, removes the auxiliary one-to-many detection head. Pretrained checkpoints retain the full training architecture and may show higher counts. ## Train diff --git a/docs/en/tasks/pose.md b/docs/en/tasks/pose.md index 565e1632..0c087744 100644 --- a/docs/en/tasks/pose.md +++ b/docs/en/tasks/pose.md @@ -58,6 +58,7 @@ Ultralytics YOLO26 pretrained Pose models are shown here. Detect, Segment and Po - **mAPval** values are for single-model single-scale on [COCO Keypoints val2017](https://cocodataset.org/) dataset.
Reproduce by `yolo val pose data=coco-pose.yaml device=0` - **Speed** averaged over COCO val images using an [Amazon EC2 P4d](https://aws.amazon.com/ec2/instance-types/p4/) instance.
Reproduce by `yolo val pose data=coco-pose.yaml batch=1 device=0|cpu` +- **Params** and **FLOPs** values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers and, for end2end models, removes the auxiliary one-to-many detection head. Pretrained checkpoints retain the full training architecture and may show higher counts. ## Train diff --git a/docs/en/tasks/segment.md b/docs/en/tasks/segment.md index 5b43ccbc..54ea14b4 100644 --- a/docs/en/tasks/segment.md +++ b/docs/en/tasks/segment.md @@ -38,6 +38,7 @@ YOLO26 pretrained Segment models are shown here. Detect, Segment and Pose models - **mAPval** values are for single-model single-scale on [COCO val2017](https://cocodataset.org/) dataset.
Reproduce by `yolo val segment data=coco.yaml device=0` - **Speed** averaged over COCO val images using an [Amazon EC2 P4d](https://aws.amazon.com/ec2/instance-types/p4/) instance.
Reproduce by `yolo val segment data=coco.yaml batch=1 device=0|cpu` +- **Params** and **FLOPs** values are for the fused model after `model.fuse()`, which merges Conv and BatchNorm layers and, for end2end models, removes the auxiliary one-to-many detection head. Pretrained checkpoints retain the full training architecture and may show higher counts. ## Train