Satellite Images Annotator for Tree Detection : 6000 Satellite Images (YOLO Format - Tree Detection)
Budget: $750 – $1,500 USD
We are looking for an experienced data annotator (or a team) to annotate 12,000 high-resolution satellite images. The goal is to train a YOLO-based deep learning model to detect trees.
Total Volume: 12,000 images.
Format: YOLO Darknet (.txt).
Class: Single class (tree).
The Workflow:
Pilot Phase: We will hire you for a paid test batch of 50 images first.
Full Project: Upon successful review of the pilot, the full contract will be awarded.
Detailed Requirements:
Bounding Boxes: Each tree visible must be outlined with a precise bounding box.
Format: One .txt file per image containing: <class_id> <x_center> <y_center> <width> <height> (normalized).
Tools: You may use CVAT, LabelImg, Roboflow, or similar. (AI-assisted labeling tools are permitted and encouraged for efficiency, provided the accuracy is verified manually).
Acceptance Criteria (Strict):
Accuracy: Missed trees or loose/overlapping boxes must be under 5% in a random sample audit.
Naming: Annotation filenames must exactly match the source image names (e.g., image_01.jpg -> image_01.txt).
Validation: Files must load without syntax errors in a standard YOLO training script.
To Apply, Please Answer:
What annotation tool do you use (CVAT, LabelImg, etc.)?
Do you use any semi-automated tools (like SAM or pre-trained models) to speed up the process?
What is your estimated turnaround time for 1,000 images?
Have you worked with aerial or satellite imagery (top-down view) before?
Total Volume: 12,000 images.
Format: YOLO Darknet (.txt).
Class: Single class (tree).
The Workflow:
Pilot Phase: We will hire you for a paid test batch of 50 images first.
Full Project: Upon successful review of the pilot, the full contract will be awarded.
Detailed Requirements:
Bounding Boxes: Each tree visible must be outlined with a precise bounding box.
Format: One .txt file per image containing: <class_id> <x_center> <y_center> <width> <height> (normalized).
Tools: You may use CVAT, LabelImg, Roboflow, or similar. (AI-assisted labeling tools are permitted and encouraged for efficiency, provided the accuracy is verified manually).
Acceptance Criteria (Strict):
Accuracy: Missed trees or loose/overlapping boxes must be under 5% in a random sample audit.
Naming: Annotation filenames must exactly match the source image names (e.g., image_01.jpg -> image_01.txt).
Validation: Files must load without syntax errors in a standard YOLO training script.
To Apply, Please Answer:
What annotation tool do you use (CVAT, LabelImg, etc.)?
Do you use any semi-automated tools (like SAM or pre-trained models) to speed up the process?
What is your estimated turnaround time for 1,000 images?
Have you worked with aerial or satellite imagery (top-down view) before?