Stats
Actions
Tags
From dora-skills
Configures YOLO object detection, SAM2 segmentation, and CoTracker point tracking nodes for dora-rs dataflow pipelines. Activates on dora-yolo, dora-sam2, and related computer vision topics.
How this skill is triggered — by the user, by Claude, or both
Slash command
/dora-skills:hub-detectionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> YOLO detection, SAM2 segmentation, and CoTracker point tracking
YOLO detection, SAM2 segmentation, and CoTracker point tracking
| Node | Install | Description | GPU Required |
|---|---|---|---|
| dora-yolo | pip install dora-yolo | YOLOv8 object detection | Recommended |
| dora-sam2 | pip install dora-sam2 | Segment Anything 2 | Required (CUDA) |
| dora-cotracker | pip install dora-cotracker | Point tracking | Recommended |
YOLOv8 object detection with bounding boxes, confidence scores, and labels.
- id: yolo
build: pip install dora-yolo
path: dora-yolo
inputs:
image: camera/image
outputs:
- bbox
env:
MODEL: yolov8n.pt # yolov5n, yolov8n/s/m/l/x
# image: UInt8Array
metadata = {
"width": 640,
"height": 480,
"encoding": "bgr8" # or "rgb8"
}
# StructArray with bounding boxes
bbox = {
"bbox": np.array([x1,y1,x2,y2, ...]).flatten(), # xyxy format
"conf": np.array([0.95, 0.87, ...]), # confidence scores
"labels": np.array(["person", "car", ...]) # class names
}
metadata = {"format": "xyxy", "primitive": "boxes2d"}
bbox_data = event["value"][0]
bbox = {
"bbox": bbox_data["bbox"].values.to_numpy().reshape(-1, 4),
"conf": bbox_data["conf"].values.to_numpy(),
"labels": bbox_data["labels"].values.to_numpy(zero_copy_only=False)
}
# Draw boxes
for i, box in enumerate(bbox["bbox"]):
x1, y1, x2, y2 = box.astype(int)
label = bbox["labels"][i]
conf = bbox["conf"][i]
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
cv2.putText(frame, f"{label} {conf:.2f}", (x1, y1-10),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
Segment Anything Model 2 for object segmentation.
- id: sam2
build: pip install dora-sam2
path: dora-sam2
inputs:
image: camera/image
bbox: yolo/bbox # Optional: use YOLO boxes as prompts
outputs:
- masks # UInt8Array segmentation masks
# masks: UInt8Array
metadata = {
"width": 640,
"height": 480,
"primitive": "masks"
}
nodes:
- id: camera
build: pip install opencv-video-capture
path: opencv-video-capture
inputs:
tick: dora/timer/millis/33
outputs:
- image
- id: yolo
build: pip install dora-yolo
path: dora-yolo
inputs:
image: camera/image
outputs:
- bbox
- id: sam2
build: pip install dora-sam2
path: dora-sam2
inputs:
image: camera/image
bbox: yolo/bbox
outputs:
- masks
- id: rerun
build: pip install dora-rerun
path: dora-rerun
inputs:
image: camera/image
detections: yolo/bbox
segmentation: sam2/masks
Real-time point tracking using Facebook's CoTracker model.
- id: tracker
build: pip install dora-cotracker
path: dora-cotracker
inputs:
image: camera/image
points_to_track: detector/points # Optional programmatic input
outputs:
- tracked_image # Visualization with tracked points
- tracked_points # Current point positions
import numpy as np
import pyarrow as pa
# Send points to track
points = np.array([
[320, 240], # Center
[160, 120], # Top-left
[480, 360] # Bottom-right
], dtype=np.float32)
node.send_output("points_to_track", pa.array(points.ravel()), {
"num_points": len(points),
"dtype": "float32",
"shape": (len(points), 2)
})
# tracked_points: Float32Array
# Same format as input points
metadata = {
"num_points": N,
"dtype": "float32",
"shape": (N, 2)
}
nodes:
- id: camera
build: pip install opencv-video-capture
path: opencv-video-capture
inputs:
tick: dora/timer/millis/100
outputs:
- image
env:
ENCODING: "rgb8"
IMAGE_WIDTH: 640
IMAGE_HEIGHT: 480
- id: yolo
build: pip install dora-yolo
path: dora-yolo
inputs:
image: camera/image
outputs:
- bbox
- centroids # Custom output: detection centers
- id: tracker
build: pip install dora-cotracker
path: dora-cotracker
inputs:
image: camera/image
points_to_track: yolo/centroids
outputs:
- tracked_image
- tracked_points
- id: rerun
build: pip install dora-rerun
path: dora-rerun
inputs:
raw_image: camera/image
tracking_viz: tracker/tracked_image
import pyarrow as pa
import numpy as np
# Create bbox structure
bbox_dict = {
"bbox": np.array([x1, y1, x2, y2, ...], dtype=np.float32),
"conf": np.array([0.95, ...], dtype=np.float32),
"labels": np.array(["person", ...])
}
# Encode as Arrow
encoded = pa.array([bbox_dict])
node.send_output("bbox", encoded, {
"format": "xyxy",
"primitive": "boxes2d"
})
# xyxy to xywh
def xyxy_to_xywh(box):
x1, y1, x2, y2 = box
return [x1, y1, x2 - x1, y2 - y1]
# xywh to xyxy
def xywh_to_xyxy(box):
x, y, w, h = box
return [x, y, x + w, y + h]
nodes:
- id: camera
build: pip install opencv-video-capture
path: opencv-video-capture
inputs:
tick: dora/timer/millis/33
outputs:
- image
env:
IMAGE_WIDTH: 640
IMAGE_HEIGHT: 480
- id: yolo
build: pip install dora-yolo
path: dora-yolo
inputs:
image: camera/image
outputs:
- bbox
env:
MODEL: yolov8n.pt
- id: rerun
build: pip install dora-rerun
path: dora-rerun
inputs:
camera_feed:
source: camera/image
metadata:
primitive: "image"
detections:
source: yolo/bbox
metadata:
primitive: "boxes2d"
env:
IMAGE_WIDTH: 640
IMAGE_HEIGHT: 480
npx claudepluginhub zhanghandong/dora-skills --plugin dora-skillsProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.