The LTCCD dataset contains 10,294 images : normal (3,350), ASCUS(1,744), ASC-H (1,600), LSIL (2,150) and HSIL (1,450). 
The resolution of each image is 128×128 pixels. These images are collected from cervical cell samples from 238 specimens. 
All specimens were produced into liquid-based preparations using Thinprep Cytological with the H&E staining method. 
All specimens were automatically scanned by a digital slide scanner in 20× objective lenses.
For raw data, the baseline (ResNet50) performance: ACC=51.07%; 
For augmented data (after partition), the baseline (ResNet50) performance: ACC=56.29%;
For augmented data (before partition), the baseline (ResNet50) performance: ACC=84.08%.
Using the LTCCD dataset, please cite: Peng Jiang, Juan Liu*, Hua Chen, Cheng Li, Baochuan Pang, Dehua Cao. 
Channel Spatial Collaborative Attention Network for Fine-grained Classification of Cervical Cells. ICONIP 2022.