czl/generated-imagenette

Generated Imagenette Dataset

Description

This repository contains the dataset used for the generative-data-augmentation project. The dataset is organized as follows:

Dataset Structure

• analysis/: This directory contains analysis related to the dataset.
• metadata/: This directory contains the list of file path used for the Synthetic (Noisy) and Synthetic (Clean) datasets.
• synthetic/: This directory contains the image files. Each folder represents a class.

Metadata Files

• metadata/synthetic-cleaned.txt: This file contains the file paths for the Synthetic (Clean) dataset.
• metadata/synthetic-noisy.txt: This file contains the file paths for the Synthetic (Noisy) dataset.

Analysis Files

• analysis/imageGen_trace_clip.csv: This file contains the trace data for the generated images, with similarity scores.
• analysis/imageGen_trace_input_clip_mean.csv: This file contains the pivot table for the mean CLIP similarity scores for the generated images by interpolation steps.
• analysis/imageGen_trace_ssim_i_avg.csv: This file contains the pivot table for the mean SSIM scores for the generated images by interpolation steps.
• analysis/imageGen_trace.csv: This file contains the trace data for the generated images, without similarity scores.
• analysis/ori_synth_regplot.svg: This file contains the regression plot for the similarity scores between the original and synthetic images.
• analysis/ssim_regplot.svg: This file contains the regression plot for the SSIM scores between the original and synthetic images.
• analysis/text_synth_regplot.svg: This file contains the regression plot for the similarity scores between the synthetic images and the text embeddings.
• analysis/val.json: This file contains the evaluation results of the baseline classifier on the validation dataset, which identified the top 5 misclassified classes for each class in the dataset.