Trans4Map / README.md
insailab's picture
docs: add Trans4Map model card
df5af83 verified
|
Raw History Blame Contribute Delete
2.85 kB
metadata
pipeline_tag: image-segmentation
library_name: pytorch
license: apache-2.0
tags:
  - vision-transformer
  - bird-eye-view
  - semantic-mapping
  - egocentric-vision
  - matterport3d
  - replica
  - pytorch
  - wacv-2023

Trans4Map Model Zoo

Official model repository for Trans4Map: Revisiting Holistic Bird's-Eye-View Mapping from Egocentric Images to Allocentric Semantics with Vision Transformers (WACV 2023).


Architecture

Trans4Map proposes an end-to-end one-stage Transformer-based framework for Bird's-Eye-View mapping:

  1. Transformer Encoder: Efficient Transformer backbone (e.g. SegFormer MiT-B2, MiT-B4) extracts rich contextual features from egocentric image batches.
  2. Bidirectional Allocentric Memory (BAM): Projects egocentric features into an allocentric memory space.
  3. Map Decoder: Parses accumulated memory and predicts the complete top-down semantic segmentation map.

Benchmark Results (Matterport3D)

Method Backbone mIoU (%) Checkpoint
ConvNeXt ConvNeXt-T 35.91 -
ConvNeXt ConvNeXt-S 36.49 -
FAN FAN-T 31.07 -
FAN FAN-S 34.62 -
Swin Swin-T 34.19 -
Swin Swin-S 36.80 -
Trans4Map MiT-B2 40.02 Tran4Map_B2_20points.pkl
Trans4Map MiT-B4 40.88 Tran4Map_B4_20points.pkl

Quick Start & Evaluation

Download Checkpoints:

# Download MiT-B2 model
hf download InSAI-Lab/Trans4Map Tran4Map_B2_20points.pkl --local-dir ./checkpoints

# Download MiT-B4 model
hf download InSAI-Lab/Trans4Map Tran4Map_B4_20points.pkl --local-dir ./checkpoints

Python Evaluation Example:

python build_test_data_feature.py
python test.py
python eval/eval.py
python eval/eval_bfscore.py

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 repository is licensed under the Apache-2.0 License.