mirror of
https://github.com/ultralytics/ultralytics
synced 2026-04-21 14:07:18 +00:00
Update Dockerfile-runner to v2.333.0 (#23974)
Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com> Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
This commit is contained in:
Glenn Jocher
GitHub
parent
4aa768eff0
commit
f514503165
3 changed files with 35 additions and 35 deletions
|
|
@ -7,7 +7,7 @@
|
|||
FROM ultralytics/ultralytics:latest
|
||||
|
||||
# Set additional environment variables for runner (no "v" on the version)
|
||||
ARG RUNNER_VERSION=2.332.0
|
||||
ARG RUNNER_VERSION=2.333.0
|
||||
ENV RUNNER_ALLOW_RUNASROOT=1 \
|
||||
DEBIAN_FRONTEND=noninteractive
|
||||
|
||||
|
|
|
|||
|
|
@ -7,7 +7,7 @@
|
|||
FROM ultralytics/ultralytics:latest-cpu
|
||||
|
||||
# Set additional environment variables for runner (no "v" on the version)
|
||||
ARG RUNNER_VERSION=2.332.0
|
||||
ARG RUNNER_VERSION=2.333.0
|
||||
ENV RUNNER_ALLOW_RUNASROOT=1 \
|
||||
DEBIAN_FRONTEND=noninteractive
|
||||
|
||||
|
|
|
|||
|
|
@ -26,19 +26,19 @@ For a deeper look at the motivation behind this architectural shift, see the [Ul
|
|||
|
||||
YOLO26 uses a **dual-head architecture** during [training](../modes/train.md). Both heads share the same backbone and neck, but produce outputs in different ways:
|
||||
|
||||
| Head | Purpose | Detection Output | Post-Processing |
|
||||
| --- | --- | --- | --- |
|
||||
| **One-to-One** (default) | End-to-end inference | `(N, 300, 6)` | Confidence threshold only |
|
||||
| **One-to-Many** | Traditional YOLO output | `(N, nc + 4, 8400)` | Requires NMS |
|
||||
| Head | Purpose | Detection Output | Post-Processing |
|
||||
| ------------------------ | ----------------------- | ------------------- | ------------------------- |
|
||||
| **One-to-One** (default) | End-to-end inference | `(N, 300, 6)` | Confidence threshold only |
|
||||
| **One-to-Many** | Traditional YOLO output | `(N, nc + 4, 8400)` | Requires NMS |
|
||||
|
||||
The shapes above are for [detection](../tasks/detect.md). Other tasks extend the one-to-one output with additional data per detection:
|
||||
|
||||
| Task | End-to-End Output | Extra Data |
|
||||
| --- | --- | --- |
|
||||
| [Detection](../tasks/detect.md) | `(N, 300, 6)` | — |
|
||||
| Task | End-to-End Output | Extra Data |
|
||||
| ----------------------------------- | ------------------------------------------ | ----------------------------------- |
|
||||
| [Detection](../tasks/detect.md) | `(N, 300, 6)` | — |
|
||||
| [Segmentation](../tasks/segment.md) | `(N, 300, 6 + nm)` + proto `(N, nm, H, W)` | `nm` mask coefficients (default 32) |
|
||||
| [Pose](../tasks/pose.md) | `(N, 300, 57)` | 17 keypoints × 3 (x, y, visibility) |
|
||||
| [OBB](../tasks/obb.md) | `(N, 300, 7)` | Rotation angle |
|
||||
| [Pose](../tasks/pose.md) | `(N, 300, 57)` | 17 keypoints × 3 (x, y, visibility) |
|
||||
| [OBB](../tasks/obb.md) | `(N, 300, 7)` | Rotation angle |
|
||||
|
||||
During training, both heads run simultaneously — the one-to-many head provides a richer learning signal, while the one-to-one head learns to produce clean, non-overlapping predictions. During [inference](../modes/predict.md) and [export](../modes/export.md), only the **one-to-one head** is active by default, producing up to 300 detections per image in the format `[x1, y1, x2, y2, confidence, class_id]`.
|
||||
|
||||
|
|
@ -74,13 +74,13 @@ When you call `model.fuse()`, it folds Conv + BatchNorm layers for faster infere
|
|||
|
||||
**Yes, the output format is different.** If you wrote custom post-processing logic for [YOLOv8](../models/yolov8.md) or [YOLO11](../models/yolo11.md) (for example, when running inference with [ONNX Runtime](../integrations/onnx.md) or [TensorRT](../integrations/tensorrt.md)), you'll need to update it to handle the new output shape:
|
||||
|
||||
| | YOLOv8 / YOLO11 | YOLO26 (end-to-end) |
|
||||
| --- | --- | --- |
|
||||
| **Detection output** | `(N, nc + 4, 8400)` | `(N, 300, 6)` |
|
||||
| **Box format** | `xywh` (center x, center y, width, height) | `xyxy` (top-left x, top-left y, bottom-right x, bottom-right y) |
|
||||
| **Layout** | Box coords + class scores per anchor | `[x1, y1, x2, y2, conf, class_id]` |
|
||||
| **NMS required** | Yes | No |
|
||||
| **Post-processing** | NMS + confidence filter | Confidence filter only |
|
||||
| | YOLOv8 / YOLO11 | YOLO26 (end-to-end) |
|
||||
| -------------------- | ------------------------------------------ | --------------------------------------------------------------- |
|
||||
| **Detection output** | `(N, nc + 4, 8400)` | `(N, 300, 6)` |
|
||||
| **Box format** | `xywh` (center x, center y, width, height) | `xyxy` (top-left x, top-left y, bottom-right x, bottom-right y) |
|
||||
| **Layout** | Box coords + class scores per anchor | `[x1, y1, x2, y2, conf, class_id]` |
|
||||
| **NMS required** | Yes | No |
|
||||
| **Post-processing** | NMS + confidence filter | Confidence filter only |
|
||||
|
||||
For [segmentation](../tasks/segment.md), [pose](../tasks/pose.md), and [OBB](../tasks/obb.md) tasks, YOLO26 appends task-specific data to each detection — see the [output shapes table](#how-end-to-end-detection-works) above.
|
||||
|
||||
|
|
@ -149,12 +149,12 @@ The following formats **do not** support end-to-end and automatically fall back
|
|||
|
||||
End-to-end detection provides significant deployment benefits with minimal impact on [accuracy](https://www.ultralytics.com/glossary/accuracy):
|
||||
|
||||
| Metric | End-to-End (default) | One-to-Many + NMS (`end2end=False`) |
|
||||
| --- | --- | --- |
|
||||
| **CPU Inference Speed** | Up to **43% faster** | Baseline |
|
||||
| **mAP Impact** | ~0.5 mAP lower | Matches or exceeds YOLO11 |
|
||||
| **Post-Processing** | Confidence filter only | Full NMS pipeline |
|
||||
| **Deployment Complexity** | Minimal | Requires NMS implementation |
|
||||
| Metric | End-to-End (default) | One-to-Many + NMS (`end2end=False`) |
|
||||
| ------------------------- | ---------------------- | ----------------------------------- |
|
||||
| **CPU Inference Speed** | Up to **43% faster** | Baseline |
|
||||
| **mAP Impact** | ~0.5 mAP lower | Matches or exceeds YOLO11 |
|
||||
| **Post-Processing** | Confidence filter only | Full NMS pipeline |
|
||||
| **Deployment Complexity** | Minimal | Requires NMS implementation |
|
||||
|
||||
For most real-world applications, the ~0.5 [mAP](https://www.ultralytics.com/glossary/mean-average-precision-map) difference is negligible, especially when considering the speed and simplicity gains. If maximum accuracy is your top priority, you can always fall back to the one-to-many head using `end2end=False`.
|
||||
|
||||
|
|
@ -197,12 +197,12 @@ Yes, that's the expected end-to-end output format for detection: [batch size](ht
|
|||
|
||||
For other tasks, the output shape differs:
|
||||
|
||||
| Task | Output Shape | Description |
|
||||
| --- | --- | --- |
|
||||
| Detection | `(1, 300, 6)` | `[x1, y1, x2, y2, conf, class_id]` |
|
||||
| Task | Output Shape | Description |
|
||||
| ------------ | ------------------------------------ | ----------------------------------------------------------------- |
|
||||
| Detection | `(1, 300, 6)` | `[x1, y1, x2, y2, conf, class_id]` |
|
||||
| Segmentation | `(1, 300, 38)` + `(1, 32, 160, 160)` | 6 box values + 32 mask coefficients, plus a prototype mask tensor |
|
||||
| Pose | `(1, 300, 57)` | 6 box values + 17 keypoints × 3 (x, y, visibility) |
|
||||
| OBB | `(1, 300, 7)` | 6 box values + 1 rotation angle |
|
||||
| Pose | `(1, 300, 57)` | 6 box values + 17 keypoints × 3 (x, y, visibility) |
|
||||
| OBB | `(1, 300, 7)` | 6 box values + 1 rotation angle |
|
||||
|
||||
### How do I check if my exported model is end-to-end?
|
||||
|
||||
|
|
@ -238,9 +238,9 @@ Yes. All YOLO26 task variants — [detection](../tasks/detect.md), [segmentation
|
|||
|
||||
Each task extends the base detection output with task-specific data:
|
||||
|
||||
| Task | Model | End-to-End Output |
|
||||
| --- | --- | --- |
|
||||
| Detection | `yolo26n.pt` | `(N, 300, 6)` |
|
||||
| Segmentation | `yolo26n-seg.pt` | `(N, 300, 38)` + proto `(N, 32, 160, 160)` |
|
||||
| Pose | `yolo26n-pose.pt` | `(N, 300, 57)` |
|
||||
| OBB | `yolo26n-obb.pt` | `(N, 300, 7)` |
|
||||
| Task | Model | End-to-End Output |
|
||||
| ------------ | ----------------- | ------------------------------------------ |
|
||||
| Detection | `yolo26n.pt` | `(N, 300, 6)` |
|
||||
| Segmentation | `yolo26n-seg.pt` | `(N, 300, 38)` + proto `(N, 32, 160, 160)` |
|
||||
| Pose | `yolo26n-pose.pt` | `(N, 300, 57)` |
|
||||
| OBB | `yolo26n-obb.pt` | `(N, 300, 7)` |
|
||||
|
|
|
|||
Loading…
Reference in a new issue