markytools/goosyntheticv3

---
dataset_info:
features:
- name: image
dtype: image
- name: split
dtype: string
- name: width
dtype: int64
- name: height
dtype: int64
- name: bboxes
dtype: string
- name: labels
dtype: string
- name: cab
dtype: int64
- name: hum
dtype: int64
- name: light
dtype: float64
- name: cam
dtype: int64
- name: env
dtype: int64
- name: gaze_item
dtype: int64
- name: gazeIdx
dtype: int64
- name: gaze_cx
dtype: int64
- name: gaze_cy
dtype: int64
- name: hx
dtype: int64
- name: hy
dtype: int64
- name: pitch
dtype: float64
- name: yaw
dtype: float64
- name: roll
dtype: float64
- name: seg
dtype: string
- name: segm_gazeIdx
dtype: int64
- name: occluded
dtype: int64
splits:
- name: train
num_bytes: 99500978350.0
num_examples: 172800
- name: test
num_bytes: 11081866319.6
num_examples: 19200
download_size: 110113558133
dataset_size: 110582844669.6
---
The dataset features/columns here are almost similar to the original github instruction (please read the github documentation first to understand the dataset): https://github.com/upeee/GOO-GAZE2021/blob/main/dataset/goosynth-download.txt

To download goosynthtrain in huggingface, run the code below (https://huggingface.co/docs/datasets/v1.10.0/loading_datasets.html#from-the-huggingface-hub):

from datasets import load_dataset</br>
dataset = load_dataset("markytools/goosyntheticv3")

The image datasets will be stored in ""~/.cache/huggingface", so you need to delete the files here if you want to free up space.

The only difference here is that there is a new feature name called "splits", ["train", "test"] </br>
The "bboxes" and "labels" features are in string format, so you can use the code below to convert the string into list:</br>
import ast</br>
listOfBboxes = ast.literal_eval(dataset["test"]["bboxes"][0])</br>
</br>
The feature "seg" is now in string format instead of numpy ndarray. This is an optional feature, and you can manually download the files here (https://huggingface.co/datasets/markytools/goosegmv3) using wget commandline. The files are in .npy so load it using np.load (https://numpy.org/doc/stable/reference/generated/numpy.load.html).

原始信息汇总

数据集概述

数据集特征

• image: 图像数据
• split: 字符串，表示数据分割类型
• width: 整数，图像宽度
• height: 整数，图像高度
• bboxes: 字符串，边界框数据
• labels: 字符串，标签数据
• cab: 整数
• hum: 整数
• light: 浮点数
• cam: 整数
• env: 整数
• gaze_item: 整数
• gazeIdx: 整数
• gaze_cx: 整数
• gaze_cy: 整数
• hx: 整数
• hy: 整数
• pitch: 浮点数
• yaw: 浮点数
• roll: 浮点数
• seg: 字符串，分割数据
• segm_gazeIdx: 整数
• occluded: 整数

数据分割

• train: 训练集，包含172800个样本，总大小为99500978350.0字节
• test: 测试集，包含19200个样本，总大小为11081866319.6字节

数据集大小

• 下载大小: 110113558133字节
• 数据集总大小: 110582844669.6字节