Add vertex POSITION REGRESSOR (DINOv2) on top of edge classifier

Upload vertex_regressor_v4_800.pt (DINOv2-based vertex position regressor)

Add DINOv2 edge classifier (v4) for post-processing edge filtering

Upload edge_classifier_v4_400.pt (DINOv2 head, 82.1% val acc)

Revert TTA: 3x inference timed out on HF Space (b6bc99a -> all zeros)

TTA: Hungarian-matched multi-seed inference with strict 3-pass agreement

Vertex view-projection refinement: snap 3D corners to 2D gestalt evidence

Orphan-vertex cleanup + apex snap + local eval harness

2D gestalt-edge consistency filter: drop edges with no view support

Add per-sample diagnostic prints (no behavior change)

Revert plane_wireframe ensemble: massive regression to 0.2412

Rollback to proven 0.4584 baseline, add plane_wireframe to ensemble

Geometric regularization pass: vertical + Manhattan-direction snap

Revert TTA + add COLMAP-support vertex filter

TTA optimization: cache point fusion, vary only priority sampling

TTA: 3-pass test-time augmentation with priority-sample seed variation

Ensemble: disable redundant heavy features in classical pipeline

Add learned + classical pipeline ensemble (LOCAL only, push tomorrow)

Lower CONF_THRESH 0.5 -> 0.4 to recover missing edges

Revert: roll back COLMAP refine + junction + snap changes

Boost: COLMAP plane refinement + junction constraints + snap fixes

Optimize hybrid geometric ensemble for massive HSS boost

Re-enable hybrid geometric tracking ensemble

Add dummy app.py to fix NO_APP_FILE Space error

Upload checkpoint.pt with huggingface_hub

Merge remote

Keep remote README.md

initial commit

Add runtime dependency installation for scipy and pandas

Temporarily disable hybrid logic for debugging

Fix OpenMP SIGABRT crash caused by pycolmap import

Fix dataset script loading and add parquet fallback

Fix submission.json output path and filter NaNs in hybrid_merge

Deploy learned baseline + hybrid multi-view tracking ensemble

Disable bundle adjust (timeout) and set TRACK_MIN_VIEWS=2 for recall

Lower sklearn edge threshold to 0.45 to boost edge recall

Enable Bundle Adjustment and Track Vertices

Add missing core Python files

Bypass validation for track edges and fix SVD crashes

Switch to alternative vertex generation strategies since DGCNN cannot run

Fix sys.path and networkx dependency for Hugging Face

Fix relative paths for PyTorch models

Optimize DGCNN edge hyper-parameters to beat baseline

Update tuned hyperparameters

Enhance wireframe prediction filtering and clustering

Fix competition config and script crash

Update team name and submission repo

submit handcrafted baseline

Update script.py

Update script.py