SynthAIzer/indian-cattle-buffalo-breeds

Indian Cattle & Buffalo Breeds - Train/Val/Test Dataset

Built from BUFFALO___CATTLE_BREEDS.pdf (Dairy Knowledge Portal / India Today).

Structure

train/<Category_Breed>/*.jpg   (2196 images, 67 classes)
val/<Category_Breed>/*.jpg     (421 images, 67 classes)
test/<Category_Breed>/*.jpg    (555 images, 67 classes)
metadata.csv / metadata.json   (breed descriptions, region, coat color, horn shape etc.)

Directly compatible with torchvision.datasets.ImageFolder.

How this was built

The source PDF only had 1-2 real photos per breed. Each original was expanded into ~25 augmented variants (rotation, flip, crop/zoom, brightness/contrast/color jitter) to create enough images to form real, non-overlapping train/val/test splits (70/15/15) per class, so the model has real data to learn from and be evaluated on.

IMPORTANT

This is a starter dataset for building and testing the full training pipeline. Because all variants of a class trace back to the same 1-2 source photos, a model trained only on this data will overfit to those specific animals' appearance and background, and will NOT generalize well to brand-new real-world uploaded photos.

To get real-world accuracy: replace/add real distinct photos per breed (multiple different individual animals, different backgrounds/lighting) into the corresponding train/<Category_Breed>/ folders before final training - the same folder names and pipeline will keep working with no code changes.