Natural‑Image Classification Datasets (Medium‑scale: 60 K – 1.3 M images)
| Dataset | Image Resolution | # Images | # Classes | Open‑source URL |
|---|---|---|---|---|
| CIFAR‑10 | 32 × 32 pixels | 60 000 images | 10 classes | huggingface.co |
| CIFAR‑100 | 32 × 32 pixels | 60 000 images | 100 classes | huggingface.co |
| ImageNet‑1K | 224 × 224 pixels | 1 281 167 images | 1 000 classes | huggingface.co |
| CINIC‑10 | 32 × 32 pixels | 270 000 images | 10 classes (same as CIFAR‑10) | huggingface.co |
| Tiny ImageNet | 64 × 64 pixels | 120 000 images (100 k train / 10 k val / 10 k test) | 200 classes | cs231n.stanford.edu |
| ImageNet‑LT | Variable (original ImageNet resolution) | ≈135 000 images (115 k train / 20 k test) | 1 000 classes | arxiv.org |
| Places‑LT | Variable (scene photos) | ≈99 000 images (62.5 k train / 36.5 k test) | 365 classes | arxiv.org |
| SUN397 | Variable (scene images) | 108 754 images | 397 classes | pytorch.org |
| SVHN | 32 × 32 pixels | ≈630 000 images (incl. extra set) | 10 classes (digits 0–9) | medium.com |
| Food‑101 | Max side 512 pixels | 101 000 images | 101 classes | pytorch.org |
| Food‑251 (iFood 2019) | Variable (web‑scraped) | 160 785 images (120 216 train / 12 170 val / 28 399 test) | 251 classes | sites.google.com |
| DeepFashion (Deep) | Variable (high‑res) | 289 222 images | 50 classes | mmlab.ie.cuhk.edu.hk |
| iNaturalist 2017 | Max side 800 pixels | 579 184 train + 95 986 val + 182 707 test (≈857 877) | 5 089 classes | openaccess.thecvf.com |
| iNaturalist 2018 | Max side 800 pixels | 437 513 train + 24 426 val + 149 394 test (≈611 333) | 8 142 classes | github.com |
| iNat Challenge 2019 | ≈800 × 800 pixels | 268 243 train + val + 35 350 test (≈303 593) | 1 010 classes | arxiv.org |
| CLEVR | 320 × 240 pixels | 100 000 images (70 k train / 15 k val / 15 k test) | Variable | cs.stanford.edu |
| Country211 | Variable | 63 300 images | 211 classes | OpenDataLab |
| MiniPlaces (Places88) | 128 × 128 pixels | 100 000 images | 88 classes | github.com |
| MSCOCO | Varied (typically ≥ 640 × 480) | 330 000 images | 80 object categories, 91 stuff categories | cocodataset.org |
| OpenImages V6 Subset | Variable | 600 000 + images | 600 + classes | kaggle.com |
| iWildCam 2021 | Varied | 203 314 train images, 60 214 test images | 206 species | iWildCam 2021 |
| fMoW | Mainly 224 × 224 pixels | >1 000 000 images | 63 classes | OpenDataLab |
| GeoDE | Mainly 224 × 224 pixels | 61 940 images | 40 classes | GeoDE |
| Google Landmarks v2 (subset) | Variable | 100 000 images | 1 000 + classes | github.com |