datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
repro-incremental-learning-of-sparse-attention-patterns-in-transformers-bundle
Reproduction bundle — Incremental Learning of Sparse Attention Patterns in Transformers
ICML 2026 paper #12503 · OpenReview vSRh1qU5sH · arXiv 2602.19143
Everything needed to re-run the reproduction whose results are recorded in the
Trackio logbook GwendalTsang/repro-incremental-learning-of-sparse-attention-patterns-in-transformers.
Layout
upstream/ the authors' official code, vendored unchanged… See the full description on the dataset page: https://huggingface.co/datasets/GwendalTsang/repro-incremental-learning-of-sparse-attention-patterns-in-transformers-bundle.Few-Shot-Class-Incremental-Learning3D_Few-Shot_Class-Incremental_Learning_Dataset
数据集说明
数据集结构
├ trainset/
├── 0/
│ ├── [label]-xxxxx.npy
│ ├── ......
├── 1/
├── ......
├── 10/
│
├ testset/
├── 0/
│ ├── xxxxx.npy
│ ├── ......
├── 1/
├── ......
├── 10/
文件夹"0"为基类数据集，文件夹"1"-"10"为新类数据集的10个增量阶段数据。每个文件夹内包含.npy格式的点云数据，可以使用numpy.load(path)读取相应文件。在trainset中，文件命名保证依照"[label]-xxxxx.npy"格式，其中第一个"-"前的内容为该点云的类别名，可以使用filename.split('-')[0]获取相应标签。
数据集读取示例
class CONTESTCIL_train(Dataset):
def __init__(self, root='./trainset', session=0… See the full description on the dataset page: https://huggingface.co/datasets/tyhuang/3D_Few-Shot_Class-Incremental_Learning_Dataset.