Padel Video & Image Annotation in Label Studio

PROJECT: Padel Sports Analytics — Manual Video/Image Labeling

We need an experienced data annotator to label padel (sport) match footage
using Label Studio. This is a structured, well-documented job with clear
instructions, example outputs, and quality bars for each task type.

You will receive a complete labeling package (~6.3 GB) containing:
- 5 independent Label Studio projects
- Pre-configured labeling interfaces (XML configs)
- All images, video clips, and pre-labels where applicable
- Step-by-step guides per project
- Example correct outputs for every project

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
SCOPE — EXACT VOLUME
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Project 1 — Shot-type classification
• 3,038 short slow-motion swing clips (venue 1 only, 4 games)
• Pre-filled with model predictions — confirm or correct
• ~15 seconds per clip

Project 2 — Player identity (P1–P4 bounding boxes)
• 276 frames with pre-drawn boxes (venue 1, 5 games)
• 252 additional frames (venue 2) — boxes must be drawn manually
• Assign each of 4 players a consistent ID per match

Project 3 — Ball position + event marking ★ LARGEST TASK
• 12,415 frames (both venues, 10 games)
• Fully manual — mark ball position AND event per frame
• No pre-labels; this is the most time-consuming project

Project 4 — Racket bounding boxes
• 916 frames (both venues)
• Draw a box around EVERY visible racket in each frame

Project 5 — Court keypoints
• 176 frames (venue 2 only)
• Place up to 12 named court keypoints per frame

TOTAL ESTIMATED EFFORT: ~150–160 hours for a careful annotator
(~19–20 full working days at 8 hrs/day)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
WHAT YOU MUST DELIVER
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

When all tasks in a project are complete:
1. Export from Label Studio as JSON (full format with annotations)
2. Name files exactly:
• 01_shot_type_export.json
• 02_player_identity_export.json
• 03_ball_and_events_export.json
• 04_racket_boxes_export.json
• 05_court_keypoints_export.json
3. Deliver all 5 files in one final submission
4. Include a short notes file listing any frames you skipped and why

We will run automatic quality checks against our gold set. If a batch
falls below the quality bar, we will return specific disagreements for
you to fix (included in scope — no extra charge for one revision round).

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
QUALITY BARS (must be met)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

• Shot-type: ~80% agreement with gold set
• Player identity: ≥95% boxes on correct person
• Ball + events: ≥85% ball within 8px; ≥85% events on correct frame
• Racket boxes: all visible rackets boxed (completeness matters most)
• Court keypoints: points within ~8px; leave invisible points blank

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
REQUIREMENTS
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

MUST HAVE:
✓ Prior experience with image/video annotation (bounding boxes, keypoints)
✓ Comfortable installing and using Label Studio locally
✓ Windows or Mac with enough disk space (~10 GB free)
✓ Stable internet for downloading the 6.3 GB package
✓ Strong attention to detail and ability to follow precise labeling rules
✓ Ability to work independently through a large dataset without supervision
✓ Good English reading comprehension (all instructions are in English)

NICE TO HAVE:
• Experience with sports video annotation
• Experience with Label Studio specifically
• Familiarity with padel or tennis (helps with shot-type judgment)
• Prior work on ball tracking or event detection datasets

NOT REQUIRED:
• Programming skills
• ML/model training knowledge

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
HOW TO APPLY
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Your proposal MUST include:
1. Examples of prior annotation work (screenshots or portfolio links)
2. Your estimated timeline to complete all 5 projects
3. Confirmation you can install Label Studio (Python 3.9+) locally
4. Your fixed price quote for the FULL scope (all 5 projects + 1 revision round)
5. Answer to screening question: "How would you handle a frame where
two teammates look identical and you cannot tell them apart?"

Proposals without annotation examples will not be considered.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
PAYMENT STRUCTURE (milestones)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Payment is milestone-based:
M1 (10%) — Setup Label Studio + complete 1 pilot game across all
applicable projects; we review before you continue
M2 (20%) — Project 1 (shot-type) complete + accepted
M3 (15%) — Project 2 (player identity) complete + accepted
M4 (40%) — Project 3 (ball + events) complete + accepted
M5 (10%) — Projects 4 + 5 complete + accepted
M6 (5%) — Final revision pass after our QA feedback

Do not start bulk labeling before M1 pilot is approved.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
WHAT WE PROVIDE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

• Complete labeling package (download link upon award)
• Label Studio XML configs (copy-paste into project settings)
• Per-project GUIDE.md with rules, hotkeys, edge cases, skip rules
• Example correct labels and expected JSON export format per project
• Reference rally video clips for context while labeling frames
• One round of QA feedback with specific items to fix

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
IMPORTANT RULES (summary)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

• Work ONE GAME AT A TIME — player identities (P1–P4) reset every match
• NEVER carry player IDs between games or between venues
• Two venues = completely different events (different players, courts)
• Where pre-labels exist: confirm or correct — don't redraw from scratch
• Ball + events: never guess a hidden ball — mark "ball_not_visible" instead
• Racket: box ALL visible rackets, not just the hitter's
• Court: leave keypoints blank if not visible — never guess off-screen

Full detailed instructions are in the package README and per-project guides.