coco-human-inpainted-objects

# About:
The dataset was collected on the https://www.rapidata.ai platform and contains tens of thousands of human annotations of 70+ different kinds of objects.
Rapidata makes it easy to collect manual labels in several data modalities with this repository containing freehand drawings on ~2000 images from the COCO dataset.
Users are shown an image and are asked to paint a class of objects with a brush tool - there is always a single such object on the image, so the task is not ambiguous.
The result of this user-interaction is a collection of lines drawn by the user on that particular image.

If you get value from this dataset and would like to see more in the future, please consider liking it.

# Challenge:
The challenge of the dataset is to aggregate the lines on each image to get an idea of where the target object is. For each image, we provide hundreds of 2D lines drawn by different humans that can be used to create bounding boxes and segmentation maps on each image of the target object. Apart from the lines, the dataset contains the COCO 2D bounding box ground truths as well as baseline predictions to beat.

# Structure:
The metadata.csv describes each image in one row:
| Column Name | Description |
|-------------------|-----------------------------------------------------------------------------|
| coco_filename | The unique identifier for each image in the COCO dataset. |
| class_name | The class/category that the user was asked to mark on the image. Same as `category_name` in COCO. |
| prediction | A baseline COCO bounding box prediction based on heatmaps. |
| ground_truth | The COCO bounding box ground truth. |
| IoU | The Intersection over Union (IoU) score between the baseline prediction and the ground truth. |
| lines | A 3D array of coordinates. Because each user can draw multiple lines, the first dimension represents different users, the second dimension represents multiple lines drawn by each user, and the third dimension represents the individual [x, y] coordinates of each line, relative to the image dimensions. |

# 关于本数据集：
本数据集采集自https://www.rapidata.ai平台，包含数万条针对70余种不同物体的人工标注数据。
Rapidata平台可便捷采集多种数据模态下的人工标注标签，本数据集包含从COCO数据集选取的约2000张图像上的手绘标注内容。
向用户展示单张图像，并要求其使用画笔工具标注指定类别的物体——每张图像中仅存在一个该类目标物体，因此标注任务无歧义。
用户交互所生成的结果，即为用户在对应图像上绘制的线条集合。
若您从本数据集获益并希望未来获取更多同类资源，不妨为其点赞支持。

# 任务挑战：
本数据集的挑战任务为：聚合单张图像上的所有标注线条，以确定目标物体的所在位置。针对每张图像，我们提供了由多名不同标注者绘制的数百条二维线条，可用于生成目标物体的边界框（bounding box）与分割掩码（segmentation map）。除标注线条外，本数据集还包含COCO数据集的二维边界框真实标注（ground truth），以及可供赶超的基线预测结果。

# 数据结构：
元数据文件metadata.csv以单行形式描述每张图像：
| 列名 | 描述 |
|-------------------|-----------------------------------------------------------------------------|
| coco_filename | COCO数据集中每张图像的唯一标识符。 |
| class_name | 要求用户在图像上标注的物体类别/分类，与COCO数据集中的`category_name`字段保持一致。 |
| prediction | 基于热力图生成的基线COCO边界框预测结果。 |
| ground_truth | COCO数据集的二维边界框真实标注。 |
| IoU | 基线预测结果与真实标注之间的交并比（Intersection over Union, IoU）得分。 |
| lines | 三维坐标数组。由于每位标注者可绘制多条线条，因此第一维度代表不同标注者，第二维度代表单标注者绘制的多条线条，第三维度代表每条线条的单个[x, y]坐标，坐标值基于图像尺寸进行归一化。