Image Dataset Collection Specialist

Tree Stump Image Dataset Collection Specialist Overview We are building a computer vision system that detects tree stumps from photographs. We need a freelancer to collect and organize a dataset of real-world tree stump images from publicly available sources. This is a data collection project only. No annotation is required. No AI-generated images are permitted. ________________________________________ Deliverables Collect: 500 Overhead Images Definition: The top surface of the stump is the primary visible surface. The camera is looking mostly downward at the stump. ________________________________________ 500 Angled Images Definition: The top surface and one or more sides of the stump are visible. ________________________________________ Objective The objective is not simply to collect 1,000 images. The objective is to collect 1,000 useful training examples that help a computer vision model learn to recognize tree stumps across many: • stump sizes • stump shapes • camera angles • lighting conditions • levels of weathering • levels of vegetation coverage • residential, commercial, and woodland settings Dataset diversity is more important than image quantity. ________________________________________ Acceptable Sources Images may be collected from: • YouTube videos • Contractor websites • Public websites • Public forums • Public articles • Public image galleries Images must be publicly accessible. ________________________________________ Acceptable Images Examples include: • Freshly cut stumps • Weathered stumps • Small stumps • Large stumps • Single stumps • Multiple stumps if clearly visible • Stumps surrounded by grass • Stumps surrounded by leaves • Stumps in wooded areas • Stumps in residential yards • Stumps in commercial properties • Stumps in construction environments ________________________________________ Unacceptable Images Do NOT include: • Standing trees • Logs • Firewood piles • Wood chip piles • Removed root balls • Drawings • Illustrations • AI-generated images • Images where the stump occupies less than 10% of the image • Extremely blurry images • Images with large watermarks • Images with significant text overlays ________________________________________ Duplicate Rules The goal is to collect a diverse dataset of tree stump images. Multiple images from the same YouTube video or website are acceptable. Multiple images of the same stump are acceptable if: • The viewing angle changes significantly • The image provides a meaningfully different training example • Different portions of the stump are visible • Environmental conditions differ substantially Do NOT include: • Exact duplicate images • Near-identical screenshots • Consecutive video frames that appear substantially the same • Images that do not provide meaningful variation for computer vision training When uncertain, include only the strongest example. ________________________________________ YouTube Image Rules Screenshots extracted from videos are acceptable. Requirements: • Capture only video content • No browser interface • No YouTube controls • No timestamps • No channel overlays • No browser windows • No visible playback controls Images should be extracted directly from the video content. ________________________________________ Image Quality Requirements Images must: • Be at least 800 pixels on the shortest side • Clearly show the stump • Be reasonably sharp • Be usable for computer vision training Accepted formats: • JPG • JPEG • PNG • WEBP ________________________________________ File Naming YouTube Images Format: YT_VideoId_FrameNumber(optional).jpg Examples: YT_dQw4w9WgXcQ_0001.jpg YT_dQw4w9WgXcQ_0002.jpg ________________________________________ Web Images Format: WEB_000001.jpg WEB_000002.jpg WEB_000003.jpg ________________________________________ Required Metadata Provide a CSV file with the following columns: Filename,SourceType,SourceURL,ViewType Example: YT_dQw4w9WgXcQ_0001.jpg,YouTube,https://youtube.com/watch?v=dQw4w9WgXcQ,Overhead WEB_000001.jpg,Web,https://example.com/image.jpg,Angled ________________________________________ Folder Structure Dataset/ Overhead/ Angled/ Metadata/ Metadata/ overhead_sources.csv angled_sources.csv ________________________________________ Review Process Before collecting the full dataset: 1. Submit 25 Overhead images 2. Submit 25 Angled images 3. Wait for approval Only after approval should the remaining dataset be collected. ________________________________________ Selection Criteria When applying, please provide: 1. Similar dataset collection experience 2. Estimated completion time 3. Confirmation that you understand the acceptance criteria Please begin your proposal with: STUMP DATASET to confirm that you have read the full description.