Expert YOLO Data Annotator Needed: QA & Correction of 7,525 Satellite Images
Budget: $250 – $750 USD
I am looking for a highly accurate and detail-oriented data annotator to review, correct, and finalize a dataset of 7,525 satellite images. This batch is part of a broader research dataset of over 12,000 images utilized to assess the Urban Green Space Index in regions like Qassim and Madinah.
The current dataset is already in YOLO format, but contains unacceptable labeling errors that must be systematically fixed. The final output must be 100% accurate, strictly formatted, and ready to plug directly into our deep learning training pipeline.
Scope of Work & Technical Requirements:
Review & Correct: Carefully examine 7,525 PNG images and their corresponding YOLO TXT annotation files. You will adjust, add, or delete bounding boxes to ensure every piece of vegetation is accurately captured.
Format: The dataset is already in YOLO format. You must maintain this standard.
Strict Naming Convention: Every image and its corresponding annotation file must have the exact same name (e.g., img_20251111094303_tile_124800_79680.png must perfectly match img_20251111094303_tile_124800_79680.txt).
Ready-to-Train: The final output must require zero post-processing or file renaming on my end.
Common Errors to Avoid (See Attached Examples):
The previous annotator made several recurring mistakes. Your job is to ensure these are completely eliminated:
Overlapping and Cluttered Boxes (e.g., Dense Palm Trees): In areas with dense vegetation, previous bounding boxes overlap heavily or group multiple trees together. Each distinct tree canopy must have its own tight, individual bounding box.
Loose Bounding Boxes: Many boxes currently capture too much dirt, shadow, or background terrain. Boxes must be tightly fitted to the visible edges of the green space/canopy.
Inconsistent Labeling: Some obvious sparse desert shrubs and trees were missed entirely, while shadows or non-vegetative dark spots were incorrectly labeled.
Quality Assurance & Payment Terms (Please Read Carefully):
We have a strict zero-tolerance policy for sloppy work or repeating the errors shown in the attached samples.
Paid Trial Milestone: You will first be assigned a paid test batch of 100 images. If this batch does not meet our accuracy threshold, the contract will end immediately.
Milestone-Based Delivery: The remaining 7,425 images will be divided into milestones.
Strict QA: Every submitted batch will undergo a spot-check. If a batch contains uncorrected overlaps, loose boxes, or mismatched PNG/TXT file names, it will be rejected.
Payment Release: Payment for each milestone will only be released once the batch passes QA and is fully compatible with our model requirements.
Ideal Candidate:
Proven experience annotating satellite or aerial imagery (specifically vegetation/trees).
Expertise in YOLO formatting and tools like CVAT, LabelImg, or Roboflow.
Flawless attention to detail.
The current dataset is already in YOLO format, but contains unacceptable labeling errors that must be systematically fixed. The final output must be 100% accurate, strictly formatted, and ready to plug directly into our deep learning training pipeline.
Scope of Work & Technical Requirements:
Review & Correct: Carefully examine 7,525 PNG images and their corresponding YOLO TXT annotation files. You will adjust, add, or delete bounding boxes to ensure every piece of vegetation is accurately captured.
Format: The dataset is already in YOLO format. You must maintain this standard.
Strict Naming Convention: Every image and its corresponding annotation file must have the exact same name (e.g., img_20251111094303_tile_124800_79680.png must perfectly match img_20251111094303_tile_124800_79680.txt).
Ready-to-Train: The final output must require zero post-processing or file renaming on my end.
Common Errors to Avoid (See Attached Examples):
The previous annotator made several recurring mistakes. Your job is to ensure these are completely eliminated:
Overlapping and Cluttered Boxes (e.g., Dense Palm Trees): In areas with dense vegetation, previous bounding boxes overlap heavily or group multiple trees together. Each distinct tree canopy must have its own tight, individual bounding box.
Loose Bounding Boxes: Many boxes currently capture too much dirt, shadow, or background terrain. Boxes must be tightly fitted to the visible edges of the green space/canopy.
Inconsistent Labeling: Some obvious sparse desert shrubs and trees were missed entirely, while shadows or non-vegetative dark spots were incorrectly labeled.
Quality Assurance & Payment Terms (Please Read Carefully):
We have a strict zero-tolerance policy for sloppy work or repeating the errors shown in the attached samples.
Paid Trial Milestone: You will first be assigned a paid test batch of 100 images. If this batch does not meet our accuracy threshold, the contract will end immediately.
Milestone-Based Delivery: The remaining 7,425 images will be divided into milestones.
Strict QA: Every submitted batch will undergo a spot-check. If a batch contains uncorrected overlaps, loose boxes, or mismatched PNG/TXT file names, it will be rejected.
Payment Release: Payment for each milestone will only be released once the batch passes QA and is fully compatible with our model requirements.
Ideal Candidate:
Proven experience annotating satellite or aerial imagery (specifically vegetation/trees).
Expertise in YOLO formatting and tools like CVAT, LabelImg, or Roboflow.
Flawless attention to detail.