The problem

Every major CV framework needs coordinates before it can run.

# YOLO 
regioncounter = RegionCounter(region=[120, 80, 640, 480])    # where does this region come from?

# SAM2/SAM3
predictor.predict(box=np.array([120, 80, 640, 480]))         # same problem

The standard workflow: open CVAT or Roboflow → grab coordinates → paste them back into code. Every. Single. Time.

The fix

import pixpick

region = pixpick.box("video.mp4", frame=10)  # drag a box on a specific video frame
zone   = pixpick.polygon("image.jpg")        # click polygon vertices

# coordinates are ready — unpack directly into any framework
# YOLO:
regioncounter = RegionCounter(
     region=zone.yolo_region,  # pass region points
     model="yolo26n.pt",
 )

# same for YOLOE
model.predict("image.jpg", visual_prompt= region.yolo_prompt)

# SAM/SAM2/SAM3:
predictor.predict(box=region.sam)

A window opens on your image, video, or a specific video frame. You interact. You get framework-ready coordinates back in Python. No round-trips.

pixpick.box() and pixpick.polygon() both accept a frame= argument when the source is a video file.

Install

pip install pixpick

Selectors

Box controls — drag to draw · R to reset · Enter to confirm · Esc to cancel

Polygon controls — LMB add point · RMB undo · Z clear · Enter confirm · Esc cancel

Line controls — LMB start → LMB end · RMB undo · Z clear · Enter confirm · Esc cancel

Output formats

Every selection object carries all the formats you'll ever need.

# ── Box ──────────────────────────────────────────────────────
region = pixpick.box("frame.jpg")

region.xyxy              # [x1, y1, x2, y2]            absolute pixels
region.xywh              # [x, y, w, h]                absolute pixels
region.norm_xywh         # [x, y, w, h]                0.0 – 1.0  ← YOLO label format
region.center            # (cx, cy)
region.area              # pixels²


# ── Polygon ───────────────────────────────────────────────────
zone = pixpick.polygon("frame.jpg")

zone.points              # [(x0,y0), (x1,y1), ...]     absolute pixels
zone.as_numpy            # np.array shape (N, 2)
zone.norm                # [(x0n,y0n), ...]             0.0 – 1.0
zone.bbox                # → Box   tight bbox around the polygon
zone.npoints             # int


## ── Line ─────────────────────────────────────────────────────
line = pixpick.line("frame.jpg")

line.points              # [(x0,y0), (x1,y1)]           absolute pixels
line.as_numpy            # np.array shape (2, 2)
line.norm                # [(x0n,y0n), (x1n,y1n)]       0.0 – 1.0
line.center              # (cx, cy)
line.length              # pixels

For more details, see Selectors.

Framework integration

Persistence

Pick once, reuse forever.

region.save("zone.json")
region = pixpick.load("zone.json")   # Box and Polygon both work

Production pattern — pick interactively the first time, load on every subsequent run:

from pathlib import Path
import pixpick

ZONE = "config/count_zone.json"

zone = pixpick.load(ZONE) if Path(ZONE).exists() else pixpick.polygon("frame.jpg")
zone.save(ZONE)

Docs

Contributing

We welcome contributions! Please open a GitHub issue or submit a pull request. For more information, see Contribution Guidelines.