We provide basic API for RH20T-P dataset, including reading multi-view/multi-frame image/depth data and corresponding annotation (ee_pos/ee_pos_2d
).
You can further develop on the basis of this API, for example, integrating them into the conversation of LLM to construct an instruction tuning dataset.
Sources | URL |
---|---|
RH20T dataset | download |
RH20T-P annotation | download |
from data.rh20tp.dataset import (
LazyRH20TPrimitiveDataset,
LazyRH20TActionDataset
)
dataset = LazyRH20TPrimitiveDataset(
data_root='/path/to/rh20t/dataset',
anno_path='/path/to/rh20tp/annotation')
Here, data_root
refers to the path of RH20T dataset, e.g., 'data/sources/RH20T'
; anno_path
refers to the path of RH20T-P annotation file, e.g., 'data/sources/rh20tp_cga_metadata_v1.0.pkl'
.