xdeng77/coconut_s

---
license: cc
dataset_info:
features:
- name: mask
dtype: image
- name: segments_info
struct:
- name: file_name
dtype: string
- name: image_id
dtype: int64
- name: segments_info
list:
- name: category_id
dtype: int64
- name: id
dtype: int64
- name: iscrowd
dtype: int64
- name: isthing
dtype: int64
- name: image_info
struct:
- name: coco_url
dtype: string
- name: date_captured
dtype: string
- name: file_name
dtype: string
- name: height
dtype: int64
- name: id
dtype: int64
- name: license
dtype: int64
- name: width
dtype: int64
splits:
- name: train
num_bytes: 874203347.2
num_examples: 118200
download_size: 817249090
dataset_size: 874203347.2
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

## How to download
1. Set up environment
```
pip install datasets tqdm
wget https://raw.githubusercontent.com/bytedance/coconut_cvpr2024/main/download_coconut.py
```

2. Use the download script to download the COCONut dataset splits.
```
python download_coconut.py --split coconut_s # default split: relabeled_coco_val, need to switch to coconut_s
```
The above command should print your download status, if you download it successfully you can see the results below:


3. Download other COCONut dataset splits.

If you want to download the other splits, you can replace the split name to "relabeled_coco_val" or "coconut_b"
NOTE: multiple splits download is not yet supported.

```
python download_coconut.py --split relabeled_coco_val --output_dir relabeled_coco_val
```

4. The mask images are nearly black as we use continuous segment ids for each image, you can use [github visualization tutorial](https://github.com/bytedance/coconut_cvpr2024/blob/main/tutorials/visualization/demo.ipynb) to create colorful masks for viewing.

Please go to our offical github repo for detailed usage instruction: https://github.com/bytedance/coconut_cvpr2024

原始信息汇总

数据集概述

数据集特征

• mask: 图像类型数据。
• segments_info: 结构化数据，包含以下字段：
  • file_name: 字符串类型。
  • image_id: 整数类型。
  • segments_info: 列表类型，包含以下字段：
    • category_id: 整数类型。
    • id: 整数类型。
    • iscrowd: 整数类型。
    • isthing: 整数类型。
• image_info: 结构化数据，包含以下字段：
  • coco_url: 字符串类型。
  • date_captured: 字符串类型。
  • file_name: 字符串类型。
  • height: 整数类型。
  • id: 整数类型。
  • license: 整数类型。
  • width: 整数类型。

数据集划分

• train:
  • num_bytes: 874203347.2 字节。
  • num_examples: 118200 个样本。

数据集大小

• download_size: 817249090 字节。
• dataset_size: 874203347.2 字节。

配置信息

• config_name: default
• data_files:
  • split: train
  • path: data/train-*