Dataset · huggan
Cleaned background-removed butterfly photo sample
huggan/smithsonian_butterflies_subset
1,000 selected images of various butterfly species with removed background for image generation.
What's inside
A filtered subset of the Smithsonian Butterflies collection prepared for training the butterfly_gan generative model. The set includes 1,000 top photographs of unique butterfly species selected via CLIP based on text descriptions (e.g., butterfly with open wings). Backgrounds are removed from all images, which include both text and pictures.
What you can do with it
- Training generative models (GAN or diffusion LoRA) on clean objects without background
- Rapid experimentation and prototyping of image generation pipelines
- Size
- 1K–10K rows
- Downloads this month
- 1.8K
- Likes
- 58
Models trained on it
- faverogian/Smithsonian128UNet 43
- ceyda/butterfly_cropped_uniq1K_512Unconditional image generation 34
- anontokio/sd-class-butterflies-32-from-zeroUnconditional image generation 31
- ceyda/ddpm-ema-butterflies-64 18
- dboshardy/ddim-butterflies-128 15
- ClassCat/ddpm-butterflies-128 14
- BartMartin/ddpm-butterflies-128 13
- bmateusz/ddpm-butterflies-128 11
How to load it
from datasets import load_dataset
ds = load_dataset("huggan/smithsonian_butterflies_subset")