Trans4Map Ground Truth Semantic Maps

Official dataset repository containing top-down semantic ground truth maps for Trans4Map: Revisiting Holistic Bird's-Eye-View Mapping from Egocentric Images to Allocentric Semantics with Vision Transformers (WACV 2023).

• GitHub Repository: InSAI-Lab/Trans4Map
• Paper: arXiv:2207.06205
• Model Checkpoints: InSAI-Lab/Trans4Map (Model Zoo)
• Collection: InSAI-Lab/Trans4Map & Egocentric Mapping

---

Dataset Overview

Trans4Map benchmarks holistic top-down bird's-eye-view (BEV) semantic mapping from egocentric observations in indoor environments (Matterport3D & Replica). This repository hosts the top-down semantic ground truth maps (.h5 files) generated across Matterport3D environments following the Semantic-MapNet (SMNet) protocol.

Directory Structure

data/
└── semmap/
    ├── 17DRP5sb8fy_0.h5
    ├── 1LXtFkjw3qL_0.h5
    ├── 1LXtFkjw3qL_1.h5
    ├── ... (87 scene ground truth .h5 files)
    └── zsNo4HB9uLZ_0.h5

Alternatively, download the full archive semmap.tar.gz.

---

Download Instructions

Using Hugging Face CLI:

# Download the consolidated tar.gz archive
hf download InSAI-Lab/Trans4Map semmap.tar.gz --type dataset --local-dir ./data

# Extract to data/semmap
tar -zxvf ./data/semmap.tar.gz -C ./data

Using Python (huggingface_hub):

from huggingface_hub import hf_hub_download
import tarfile

tar_path = hf_hub_download(
    repo_id="InSAI-Lab/Trans4Map",
    filename="semmap.tar.gz",
    repo_type="dataset",
    local_dir="./data"
)
with tarfile.open(tar_path, "r:gz") as tar:
    tar.extractall("./data")

---

Citation

@inproceedings{chen2023trans4map,
  title={Trans4Map: Revisiting Holistic Bird's-Eye-View Mapping from Egocentric Images to Allocentric Semantics with Vision Transformers},
  author={Chen, Chang and Zhang, Jiaming and Yang, Kailun and Peng, Kunyu and Stiefelhagen, Rainer},
  booktitle={2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2023}
}

License

This dataset is released under the Apache-2.0 License.