Garbage Image Annotation Project

# Waste Dataset Annotation Review / Correction Task This is a review and correction task, not full annotation from scratch. Most trash objects are already annotated. The main task is to check existing annotations, fix mistakes, and add missing objects where needed. Estimated workload: around 1.5 hours, depending on the final number of images and corrections required. ## Tool and Format Please use CVAT or any similar annotation tool that supports polygon segmentation. Export format: standard CVAT XML, COCO JSON, or another valid JSON/XML segmentation format. Keep original image file names unchanged. ## Annotation Type Use polygon segmentation. Polygons must follow the visible object contour accurately. Do not use rough bounding boxes or large approximate areas. Annotate only the visible part of each object. Do not guess hidden parts. ## Classes Use only these classes: * dumpster — regular waste containers / trash bins * large-dumpster — large industrial or oversized waste containers * dry-waste-dumpster — dry-waste containers, only if visually clear * trash — general loose garbage, bags, wrappers, receipts, paper, mixed waste, small trash on the ground * wheels — wheels, tires, visible wheel parts * boxes — cardboard boxes or box-like cardboard waste * furniture — discarded furniture or furniture parts * black — uncertain objects that may be trash but are unclear ## What Needs to Be Done For each image: 1. Review existing annotations. 2. Keep correct annotations as they are. 3. Add missing dumpsters and containers. 4. Add missing visible trash on the ground. 5. Correct wrong classes. 6. Fix inaccurate polygons. 7. Remove duplicate annotations. 8. Flag unclear cases if the object cannot be confidently classified. The main priority is missing dumpsters/containers. General trash is mostly already annotated, but please check it and add missing trash where clearly visible. ## Class Rules Use the most specific visible class. Examples: * regular trash bin → dumpster * large metal container → large-dumpster * clearly marked dry-waste container → dry-waste-dumpster * plastic bag, receipt, wrapper, mixed garbage → trash * cardboard box → boxes * wheel or tire → wheels * discarded chair, table, mattress, shelf → furniture * unclear dark object that may be trash, dirt, leaf, or shadow → black Do not use black for clearly visible trash. If it is clearly garbage, use trash. Do not label boxes, wheels, or furniture as trash if the specific class is clear. ## Polygon Quality Rules * Follow the object contour closely. * Do not include background, pavement, walls, grass, or shadows. * Do not include empty space around the object. * Separate objects should be annotated separately. * Overlapping objects should be annotated only by their visible parts. * Wheels should be annotated separately from dumpsters if visible. * If there is non-trash inside a trash area, exclude it where possible. * Small ground trash should be annotated if it is clear enough to identify. ## Deliverables Please provide: 1. One annotation file per image. 2. Export in valid JSON or XML format. 3. Original image file names preserved. 4. Short completion report with: * number of images reviewed * number of images corrected * unclear/ambiguous cases * any images with poor visibility, blur, or occlusion ## Acceptance Criteria The task is complete when: * all visible dumpsters/containers are annotated * missing visible trash on the ground is added * existing annotations are checked and corrected where needed * every object is labeled once and only once * polygons are accurate and clean * class names are used consistently * exported files are valid for a computer-vision pipeline

Project ID: 40518673