Release 0.1.0: xcalib model weights

README.md

@@ -0,0 +1,65 @@
+---
+library_name: xcalib
+tags:
+- camera-lidar
+- cross-modal-matching
+- extrinsic-calibration
+- jetson
+- onnx
+- tensorrt
+pipeline_tag: keypoint-detection
+---
+
+# xcalib model weights
+
+Trained checkpoints for the matching front-end studied in
+*Position Encoding in Detection-Based LiDAR-Camera Matching: A Diagnostic Study at Infrastructure Sites*.
+
+This repo is public and accompanies the accepted xcalib paper.
+
+## Files
+
+| model | site | checkpoint | config | SHA-256 |
+|---|---|---|---|---|
+| `calibrefine` | `a9_dataset_r02_s01` | `checkpoints/calibrefine_a9_dataset_r02_s01_best.pth` | `configs/calibrefine_a9_dataset_r02_s01.yaml` | `768cce41b221c64136060e26c9573f04124c839a47397aef30de6a1e9152efe1` |
+| `crlite` | `a9_dataset_r02_s01` | `checkpoints/crlite_a9_dataset_r02_s01_best.pth` | `configs/crlite_a9_dataset_r02_s01.yaml` | `16a07ed95befb632c67fe48ac3b62bdb342e21ec26072649b1acb556e0bd5dd8` |
+| `crlite_2dpe` | `a9_dataset_r02_s01` | `checkpoints/crlite_2dpe_a9_dataset_r02_s01_best.pth` | `configs/crlite_2dpe_a9_dataset_r02_s01.yaml` | `c0c523d0fc945432fb9ef03132774bed2508ade91512c4a699b2d8201103a920` |
+| `crlite_vit_exp1` | `a9_dataset_r02_s01` | `checkpoints/crlite_vit_exp1_a9_dataset_r02_s01_best.pth` | `configs/crlite_vit_exp1_a9_dataset_r02_s01.yaml` | `66d0cd7a7c1d88d8415da844798ea978d415c184409989b9a2f13a0061c57222` |
+| `crlite_vit_exp3` | `a9_dataset_r02_s01` | `checkpoints/crlite_vit_exp3_a9_dataset_r02_s01_best.pth` | `configs/crlite_vit_exp3_a9_dataset_r02_s01.yaml` | `faa7cc851cb11ac748657edf4bfb03d9988d108e936c3ba46950ec734d48358d` |
+
+## Documentation
+
+Documentation (installation, API reference, input protocol, ONNX/TensorRT
+export) is hosted at **https://radar-lab.github.io/xcalib**:
+
+- [Home & quickstart](https://radar-lab.github.io/xcalib/)
+- [Input protocol](https://radar-lab.github.io/xcalib/protocol/)
+- [Models & datasets - loading from the Hub](https://radar-lab.github.io/xcalib/hub/)
+- [API reference](https://radar-lab.github.io/xcalib/api/)
+
+## Authors
+
+Lihao Guo, Jiahao Tang, Tam Bang, Tianya Zhang, Austin Harris, Mina Sartipi, Siyang Cao
+
+## Usage
+
+```python
+from xcalib import Matcher
+
+matcher = Matcher.from_pretrained("crlite", site="a9_dataset_r02_s01")
+```
+
+## Citation
+
+```bibtex
+@article{guo2026xcalib,
+  author  = {Guo, Lihao and Tang, Jiahao and Bang, Tam and Zhang, Tianya and
+             Harris, Austin and Sartipi, Mina and Cao, Siyang},
+  title   = {Position Encoding in Detection-Based LiDAR--Camera Matching:
+             A Diagnostic Study at Infrastructure Sites},
+  journal = {IEEE Sensors Letters},
+  year    = {2026},
+  note    = {Accepted. Paper URL pending. Code:
+             https://github.com/radar-lab/xcalib},
+}
+```

checkpoints/calibrefine_a9_dataset_r02_s01_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:768cce41b221c64136060e26c9573f04124c839a47397aef30de6a1e9152efe1
+size 216974633

checkpoints/crlite_2dpe_a9_dataset_r02_s01_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0c523d0fc945432fb9ef03132774bed2508ade91512c4a699b2d8201103a920
+size 25557723

checkpoints/crlite_a9_dataset_r02_s01_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16a07ed95befb632c67fe48ac3b62bdb342e21ec26072649b1acb556e0bd5dd8
+size 25627363

checkpoints/crlite_vit_exp1_a9_dataset_r02_s01_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66d0cd7a7c1d88d8415da844798ea978d415c184409989b9a2f13a0061c57222
+size 63854676

checkpoints/crlite_vit_exp3_a9_dataset_r02_s01_best.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:faa7cc851cb11ac748657edf4bfb03d9988d108e936c3ba46950ec734d48358d
+size 65878856

configs/calibrefine_a9_dataset_r02_s01.yaml

@@ -0,0 +1,10 @@
+# CalibRefine (Pairwise baseline: Bottleneck ResNet + PointNet2 + 2D PE) — A9 r02_s01.
+model: calibrefine
+device: auto
+num_classes: 10
+crop_size: 32
+point_cloud_size: 1024
+token_len: 256
+num_blocks: [2, 2, 2, 2]
+bbox_expansion: 1.25
+weights_path: checkpoints/calibrefine_a9_dataset_r02_s01_best.pth

configs/crlite_2dpe_a9_dataset_r02_s01.yaml

@@ -0,0 +1,13 @@
+# CRLite-2DPE (ResNet + only 2D sinusoidal PE branch, R1 #6 ablation) — A9 r02_s01.
+model: crlite_2dpe
+device: auto
+num_classes: 10
+crop_size: 32
+point_cloud_size: 1024
+embed_dim: 256
+token_len: 256
+top_k: 10
+max_depth: 100.0
+pe_mode: 2d_only         # forced by CRLite2DPEModel; left here for visibility
+bbox_expansion: 1.25
+weights_path: checkpoints/crlite_2dpe_a9_dataset_r02_s01_best.pth

configs/crlite_a9_dataset_r02_s01.yaml

@@ -0,0 +1,13 @@
+# CRLite (Hybrid two-stage + Enhanced 3D PE) — A9 r02_s01 weights.
+model: crlite
+device: auto
+num_classes: 10
+crop_size: 32
+point_cloud_size: 1024
+embed_dim: 256
+token_len: 256
+top_k: 10
+max_depth: 100.0
+pe_mode: full
+bbox_expansion: 1.25
+weights_path: checkpoints/crlite_a9_dataset_r02_s01_best.pth

configs/crlite_vit_exp1_a9_dataset_r02_s01.yaml

@@ -0,0 +1,19 @@
+# CRLite-ViT Exp1 (Crop ViT + PointNet, no PE) — A9 r02_s01.
+model: crlite_vit_exp1
+device: auto
+num_classes: 10
+crop_size: 64
+point_cloud_size: 1024
+embed_dim: 256
+top_k: 10
+vit:
+  patch_size: 8
+  embed_dim: 256
+  depth: 6
+  num_heads: 8
+  mlp_ratio: 4.0
+  dropout: 0.0
+  attention_dropout: 0.0
+  pool: cls
+bbox_expansion: 1.25
+weights_path: checkpoints/crlite_vit_exp1_a9_dataset_r02_s01_best.pth

configs/crlite_vit_exp3_a9_dataset_r02_s01.yaml

@@ -0,0 +1,21 @@
+# CRLite-ViT Exp3 (Crop ViT + PointNet + Enhanced 3D PE) — A9 r02_s01.
+model: crlite_vit_exp3
+device: auto
+num_classes: 10
+crop_size: 64
+point_cloud_size: 1024
+embed_dim: 256
+token_len: 256
+max_depth: 100.0
+top_k: 10
+vit:
+  patch_size: 8
+  embed_dim: 256
+  depth: 6
+  num_heads: 8
+  mlp_ratio: 4.0
+  dropout: 0.0
+  attention_dropout: 0.0
+  pool: cls
+bbox_expansion: 1.25
+weights_path: checkpoints/crlite_vit_exp3_a9_dataset_r02_s01_best.pth