ddecosmo/ScenicOrNot_224_fine-tune_rev2

license: cc-by-sa-4.0 task_categories:

• —image-classification
• —regression tags:
• —scenicornot
• —geograph
• —landscape
• —crowdsourced size_categories:
• —10K<n<100K ---

Dataset Card for ScenicOrNot 224 Fine-Tune (Rev 2)

Dataset Summary

This dataset is a preprocessed, fine-tuning-ready version of the ScenicOrNot dataset, designed for training machine learning models on landscape aesthetics. The images in this version have been standardized to a 224x224 resolution, making them directly compatible with standard vision models (such as ResNet, MobileNet, and ViT) that expect a 1:1 aspect ratio.

The dataset includes the corresponding crowdsourced scenicness ratings, geographical coordinates, and variance data for each image.

Supported Tasks

• —Image Classification / Regression: Predicting the aesthetic "scenicness" score (1.0 to 10.0) of an outdoor image.
• —Geospatial Analysis: Correlating visual landscape features with geographical coordinates.

Dataset Structure

Splits

The data has been partitioned into three distinct splits for model training and evaluation:

• —train: 25,500 rows
• —validation: 8,500 rows
• —test: 8,500 rows

Data Fields

• —id (int64): Unique internal identifier for the dataset row.
• —gridimage_id (int64): The specific Geograph URI/identifier for the image.
• —lat (float64): The latitude coordinate of where the image was captured.
• —lon (float64): The longitude coordinate of where the image was captured.
• —average (float64): The mean scenic score calculated from the crowdsourced votes.
• —variance (float64): The population variance of the submitted votes.
• —votes (string): A comma-separated list of the individual integer scores submitted by users.
• —title (string): The title or brief description of the image as provided on Geograph.
• —imagetaken (string): The date the image was taken.
• —image (image): The photograph, resized and cropped to exactly 224x224 pixels.
• —__index_level_0__ (int64): The original index level inherited from the source dataframe.

Source Data

Images are sourced from Geograph Britain and Ireland. The crowdsourced rating platform, Scenic-Or-Not, was originally created by mySociety and is currently maintained by the Data Science Lab at Warwick Business School.

Licensing and Attribution

This dataset is distributed under a Creative Commons Attribution-ShareAlike (CC BY-SA 4.0) license.

• —Images and Metadata: Licensed by Geograph UK contributors.
• —Scenic Ratings: Licensed by the Data Science Lab UK / mySociety under the Open Database License. When using this dataset, appropriate credit must be given to the original photographers via Geograph and the Data Science Lab UK.