Skip to content

data.md

Latest commit

 

History

History
36 lines (27 loc) · 3.51 KB

data.md

File metadata and controls

36 lines (27 loc) · 3.51 KB
Raw

Data and Downloads

Download files for all provided data and pretrained models and results are listed below. All the files are in the following folder: Drive Folder

Data for finetuning the 2D segmentation model

This data uses for fine-tuning the 2D segmentation model CLIPSeg using the Wikiscenes dataset. (more about it here: distillation_adaptation.md)

  • WikiScenes (cathedrals+mosques) image data:
    • Available from WikiScenes repo (high-res 1200px images)
    • Store these in data/wikiscenes/cathedrals and data/wikiscenes/mosques
  • WikiScenes (+mosques) metadata: metadata.csv
  • Distilled semantic pseudo-labels pseudolabels.csv for WikiScenes (+mosques)
  • Metadata files: seg_crop_metadata.csv, seg_hash_metadata.csv. extract to data/ folder.
  • Correspondence-based data: hashdata.tar.gz; extract to data/hashdata

Test scene data

  • Test scene images and COLMAP reconstructions: rgb_data.tar.gz.
    (This dataset uses for training the HaLo-NeRF, more about it here: 3d_localization.md).
  • GT semantic masks: HolyScenes.tar.gz (It uses for evaluating the 2D segmentation model and 3D localization).
  • RGB reconstructions: rgb_reconstructions.tar.gz (It uses for retrieval (in run_retrieval.py, more about it here: 3d_localization.md).

Pretrained models

  • Fine-tuned CLIP model clip_ft.tar.gz (It uses for fine-tuning the 2D segmentation model here: finetune_seg.py distillation_adaptation.md)
  • Fine-tuned Segmentation model: clipseg_ft.tar.gz (It uses for the 3D localization. More about it here: 3d_localization.md).
  • Trained RGB NeRF models: rgb_models.tar.gz (The trained RGB NeRF models after the train_rgb.py part in here 3d_localization.md).
  • Trained Semantic NeRF models: semantic_models.tar.gz (The trained RGB NeRF models after the train_rgb.py part in here 3d_localization.md).

Inference results

  • Retrieval data: retrieval.tar.gz (The results of the retrieval part run_retrieval.py in here: 3d_localization.md).
  • Scene semantic segmentation data: semantic_data.tar.gz (The results of the 2D semantic segmentation part run_segmentation.py in here: 3d_localization.md).
  • RGB NeRF results: nerf_rgb.tar.gz (The results of the RGB prediction of the HaLo-NeRF: using the eval.py code. See more here: evaluation.md).
  • Semantic HaLo-NeRF results: semantic_results.tar.gz (The results of the semantic prediction of the HaLo-NeRF: using the eval.py code. See more here: evaluation.md).