Hi, and thanks for releasing the code.
I am a PhD student working on holistic 3D scene understanding from a single panorama, and I
would like to compare my method against PanoContext-Former under your exact protocol. A few
things seem to be missing for that:
eval_layout_mesh_detector_igibson.py points to
log/panocontextformer_0114_F2/igibson_1673696403/ckpt_epoch_last.pth, which is not in the
repo. Is a trained checkpoint available anywhere?
- The Data section of the README is still marked TODO. Could the rendered iGibson-Synthetic
panoramas, or the render config plus the train/test split, be shared?
- There are several
train_detection_layoutmesh_*.py variants. Is _0112_cube_phy.py the
one that reproduces the paper's numbers?
Happy to contribute a setup guide back once I get it running. Thanks!
Hi, and thanks for releasing the code.
I am a PhD student working on holistic 3D scene understanding from a single panorama, and I
would like to compare my method against PanoContext-Former under your exact protocol. A few
things seem to be missing for that:
eval_layout_mesh_detector_igibson.pypoints tolog/panocontextformer_0114_F2/igibson_1673696403/ckpt_epoch_last.pth, which is not in therepo. Is a trained checkpoint available anywhere?
panoramas, or the render config plus the train/test split, be shared?
train_detection_layoutmesh_*.pyvariants. Is_0112_cube_phy.pytheone that reproduces the paper's numbers?
Happy to contribute a setup guide back once I get it running. Thanks!