[Dataset] RSSCN7 and RESISC45

[Hazel-Heejeong-Nam]

What does this PR do?

This PR adds two new dataset to stable-datasets:

• RSSCN7
• RESISC45

Usage Examples

• RSSCN7

from stable_datasets.images.rsscn7 import RSSCN7

print("Loading RSSCN7 dataset...")
rsscn7_train = RSSCN7(split="train")
rsscn7_all = RSSCN7(split=None)

print(f"\nDataset Metadata:")
print(f"  - Homepage: {rsscn7_train.info.homepage}")
print(f"  - Description: {rsscn7_train.info.description}")
print(f"  - Citation:\n{rsscn7_train.info.citation}")

print(f"\nDataset Statistics:")
print(f"  - Train samples: {len(rsscn7_train)}")
print(f"  - Total splits: {len(rsscn7_all)}")
print(f"  - Number of classes: {rsscn7_train.features['label'].num_classes}")

sample = rsscn7_train[0]
print(f"\nSample Information:")
print(f"  - Keys: {list(sample.keys())}")
print(f"  - Image type: {type(sample['image'])}")
print(f"  - Image size: {sample['image'].size}")
print(f"  - Label (int): {sample['label']}")
print(f"  - Label (string): {rsscn7_train.features['label'].int2str(sample['label'])}")

print(f"\nAll class names:")
for i in range(7):
    print(f"  {i}: {rsscn7_train.features['label'].names[i]}")

print("\nRSSCN7 dataset loaded successfully!")

• RESISC45

from stable_datasets.images.resisc45 import RESISC45

print("Loading RESISC45 dataset...")
resisc45_train = RESISC45(split="train")
resisc45_all = RESISC45(split=None)

print(f"\nDataset Metadata:")
print(f"  - Homepage: {resisc45_train.info.homepage}")
print(f"  - Description: {resisc45_train.info.description}")
print(f"  - Citation:\n{resisc45_train.info.citation}")

print(f"\nDataset Statistics:")
print(f"  - Train samples: {len(resisc45_train)}")
print(f"  - Total splits: {len(resisc45_all)}")
print(f"  - Number of classes: {resisc45_train.features['label'].num_classes}")

sample = resisc45_train[0]
print(f"\nSample Information:")
print(f"  - Keys: {list(sample.keys())}")
print(f"  - Image type: {type(sample['image'])}")
print(f"  - Image size: {sample['image'].size}")
print(f"  - Label (int): {sample['label']}")
print(f"  - Label (string): {resisc45_train.features['label'].int2str(sample['label'])}")

print(f"\nFirst 10 class names:")
for i in range(10):
    print(f"  {i}: {resisc45_train.features['label'].names[i]}")

print("\nRESISC45 dataset loaded successfully!")

Who can review?

@RandallBalestriero

[Leon-Leyang]

Could you run pre-commit run --all-files so that the precommit test passes.

[Leon-Leyang]

My guess about the test failure is that the download is unsuccessful. Could you also try with the new download utility. So could you try just using _split_generators from BaseDatasetBuilder instead of using the current one defined in RESISC45.

[Hazel-Heejeong-Nam]

@Leon-Leyang

• Pre-commit checks are passing now.
• The failing tests are unrelated to RESISC45 and they are coming from other datasets as you can check in the log.

---

• FYI, I also changed the download link for RESISC45. The original Figshare URL used to work, but when re-testing it today, the request is now being blocked as bot traffic. A curl -IL on the original link returns:

x-amzn-waf-action: challenge
server: awselb/2.0
content-length: 0