Dataset-and-Model/Dataset-TransferLearning

Dataset Card: Taiwan Wildlife Imagery Dataset(Transfer Learning)

Dataset Summary

This dataset is primarily designed for training and evaluating Taiwan wildlife image classification models. The total storage size is approximately 7.95GB, comprising wildlife monitoring imagery captured via infrared trigger cameras across various altitudinal gradients in Taiwan.

The dataset focuses on medium-to-large terrestrial mammals (e.g., Reeves's Muntjac, Formosan Macaque) and ground-dwelling birds. To fully comply with information security protocols and Double-blind Review principles, all precise geographical coordinates have been generalized. Furthermore, any accidental human captures, patrol footprints, or proprietary equipment identifiers have been strictly filtered out to ensure complete privacy and metadata anonymity. This dataset operates independently and has no association or data linkage with any specific school, academic institution, or commercial enterprise.

Dataset Structure

• 0: Civet
• 1: Emerald Dove
• 2: Formosan Gem-faced Civet
• 3: Formosan Ferret-badger
• 4: Formosan Sambar Deer
• 5: Formosan Serow
• 6: Formosan Sika Deer
• 7: Formosan Yellow-throated Marten
• 8: Formosan Macaque
• 9: Formosan Wild Boar
• 10: Mouse
• 11: Reeves's Muntjac
• 12: Swinhoe's Pheasant
• 13: Taiwan Partridge
• 14: Urva
• 15: Cat
• 16: Dog
• 17: Squirrel

File Structure

.
├── train/         # trainset data.
  ├──00(Civet)
  ├──01(Emerald Dove)
  ├──02(Formosan Gem-faced Civet)
  .
  .
  .
  └──17(Squirrel)
└── test/          # testset data.
  ├──00(Civet)
  ├──01(Emerald Dove)
  ├──02(Formosan Gem-faced Civet)
  .
  .
  .
  └──17(Squirrel)

Data Collection & Preprocessing

Data Sources

• This dataset was legally collected by an independent research group through field operations or integrated and provided by anonymous ecological volunteers under strict de-identification conditions.
• Disclaimer: This dataset does not sync, link, or associate with the internal institutional data of any specific university, academic department, or commercial entity.

Data Anonymization

To maintain rigorous information security, data privacy, and academic double-blindness, the following de-identification procedures have been fully executed:

• Direct Identifier Removal: Personal information such as names, student IDs, national ID numbers, emails, phone numbers, and IP addresses have been completely removed.

Limitations & License

• This dataset is distributed under the [MIT License].
• Users of this dataset must strictly adhere to personal privacy and academic ethics principles. It is strictly prohibited to maliciously cross-reference image backgrounds, residual data patterns, or cross-regional clues to reverse-engineer or identify specific individuals, research groups, or academic institutions. Users shall bear full academic and legal liabilities for any such improper usage.