Add files using upload-large-folder tool

README.md

@@ -1,3 +1,217 @@
 ---
 license: cc-by-nc-sa-3.0
+task_categories:
+- object-detection
+tags:
+- lidar
+- 3d-object-detection
+- autonomous-driving
+- adversarial-robustness
+- point-cloud
+- kitti
+size_categories:
+- 10K<n<100K
+configs:
+- config_name: inject_easy
+  data_files:
+  - split: train
+    path: data/inject_easy/train-*
+- config_name: inject_medium
+  data_files:
+  - split: train
+    path: data/inject_medium/train-*
+- config_name: inject_hard
+  data_files:
+  - split: train
+    path: data/inject_hard/train-*
+- config_name: removal_az10
+  data_files:
+  - split: train
+    path: data/removal_az10/train-*
+- config_name: removal_az20
+  data_files:
+  - split: train
+    path: data/removal_az20/train-*
+- config_name: removal_az30
+  data_files:
+  - split: train
+    path: data/removal_az30/train-*
+- config_name: removal_az40
+  data_files:
+  - split: train
+    path: data/removal_az40/train-*
+- config_name: removal_az50
+  data_files:
+  - split: train
+    path: data/removal_az50/train-*
+- config_name: removal_az60
+  data_files:
+  - split: train
+    path: data/removal_az60/train-*
 ---
+
+# ATLAS-KITTI
+
+Adversarial LiDAR point clouds for evaluating the robustness of 3D object
+detectors, generated by applying **physically-grounded spoofing attacks** to the
+KITTI 3D object detection validation split.
+
+Each config is a complete copy of the KITTI val set with one attack applied to
+every frame. Clouds are **self-contained** — no local KITTI copy is required — and
+each frame ships both the attack annotation and the clean KITTI ground truth, so
+the data supports attack-success-rate *and* standard detection metrics.
+
+> **Anonymous release.** This dataset is published anonymously to support the peer
+> review of a submission under review. Author and affiliation details are withheld
+> for the review period.
+
+Companion dataset: **ATLAS-nuScenes** (planned; not yet released).
+
+## Attacks
+
+| family | configs | what it does |
+|---|---|---|
+| **Injection** | `inject_{easy,medium,hard}` | Adds a *phantom vehicle* that does not exist. Success = the detector reports an object overlapping `spoof_gt`. `easy`/`medium`/`hard` are decreasing phantom point densities (denser traces are easier to detect). |
+| **Removal** | `removal_az{10,20,30,40,50,60}` | Deletes points in an azimuth wedge aimed at a real vehicle, hiding it. Success = the detector *fails* to report the object at `spoof_gt`. The suffix is the wedge width in degrees. |
+
+Both are occlusion-consistent: injected phantoms are raycast against the scene, so
+points that would fall behind existing geometry are removed, and real points the
+phantom occludes are deleted. Phantom intensity is resampled from range-conditioned
+statistics measured on real KITTI returns.
+
+Removal follows the A-HFR (Adaptive High-Frequency Removal) model, with per-width
+removal probabilities calibrated against the physical measurements in Cao et al.,
+*"You Can't See Me: Physical Removal Attacks on LiDAR-based Autonomous Vehicles
+Driving Frameworks"* (USENIX Security 2023).
+
+## ⚠️ Frame counts differ per config — read this before computing metrics
+
+Frames that could not be attacked were **not written**. Every frame present in a
+config *is* attacked; there are no clean frames.
+
+| config | frames | absent | reason |
+|---|---|---|---|
+| `inject_easy` | 3769 | 0 | the full val split |
+| `inject_medium` | 3767 | 2 | phantom entirely removed by occlusion |
+| `inject_hard` | 3649 | 120 | sparsest traces, most vulnerable to occlusion |
+| `removal_az10`…`az60` | 3384 | 385 | frame contains no suitable vehicle to erase |
+
+**The ASR denominator is the config's own frame count, not 3769.** Dividing by 3769
+understates removal ASR by ~10%. The 385 absent removal frames are identical across
+all six widths (same target-selection criteria).
+
+## Usage
+
+```python
+from datasets import load_dataset
+import numpy as np
+
+ds = load_dataset("ps3020/atlas-kitti", "removal_az20", split="train")
+ex = ds[0]
+
+# points are stored FLAT; reshape to recover the cloud
+pts = np.asarray(ex["points"], np.float32).reshape(ex["num_points"], ex["point_dim"])
+# -> (N, 4) = x, y, z, intensity
+
+attacked_box = np.asarray(ex["spoof_gt"], np.float32)          # (7,)
+clean_boxes  = np.asarray(ex["gt_boxes"], np.float32).reshape(ex["num_gt"], 7)
+clean_names  = ex["gt_names"]
+```
+
+Computing ASR:
+
+```python
+n_success = 0
+for ex in ds:
+    pred = my_detector(reshape_points(ex))
+    hit = overlaps(pred, ex["spoof_gt"], iou_thresh=0.3)
+    # injection: success = a detection appeared;  removal: success = it vanished
+    n_success += hit if "inject" in config else (not hit)
+asr = n_success / len(ds)     # NOT / 3769
+```
+
+## Fields
+
+Common to all configs:
+
+| field | type | description |
+|---|---|---|
+| `frame_id` | string | KITTI frame id, e.g. `"000001"` |
+| `points` | list[float32] | attacked cloud, **flattened**; reshape with `num_points`/`point_dim` |
+| `num_points`, `point_dim` | int | cloud shape (`point_dim` = 4: x, y, z, intensity) |
+| `spoof_gt` | list[float32] | (7,) the attacked object: phantom box (injection) or erased box (removal), as `[x, y, z, dx, dy, dz, heading]` in LiDAR coords |
+| `gt_boxes` | list[float32] | clean KITTI GT, flattened `(num_gt, 7)`; `DontCare` excluded |
+| `gt_names` | list[string] | class per box, aligned with `gt_boxes` |
+| `num_gt` | int | object count |
+| `gt_difficulty`, `gt_num_points` | list[int] | KITTI difficulty, points-in-box |
+
+Injection configs add:
+
+| field | description |
+|---|---|
+| `atk_perturbation` | `"injection"` |
+| `atk_offset`, `atk_yaw` | phantom placement (x, y) and heading |
+| `atk_n_spoof_r` / `atk_n_spoof_k` | phantom points before / after occlusion filtering |
+| `atk_seed` | RNG seed |
+
+Removal configs add:
+
+| field | description |
+|---|---|
+| `atk_az_center_rad`, `atk_az_width_deg` | wedge bearing and width |
+| `atk_el_center_rad`, `atk_el_width_deg` | wedge elevation extent |
+| `atk_p_remove` | per-point removal probability for this width |
+| `atk_n_points_before` / `_in_sector` / `_removed` / `_after` | exact point accounting |
+| `atk_target_box`, `atk_target_range_m` | victim vehicle box and range |
+| `atk_active_this_frame` | whether the A-HFR firing gate fired |
+| `atk_range_gate_passed`, `atk_extrapolated_close_range`, `atk_inactive_reason` | gate diagnostics |
+| `atk_mode`, `atk_max_az_step_deg`, `atk_fix_elevation`, `atk_raw_target_*` | attack config |
+
+The realised removal rate for a frame is
+`atk_n_points_removed / atk_n_points_in_sector`.
+
+## Notes and limitations
+
+- **KITTI 3D object has no sequences**, so there is no temporal dimension here:
+  every frame is attacked independently, and `global` vs `ego-relative` phantom
+  placement are equivalent (hence one injection family, not two).
+- Clouds are **camera-FOV cropped** (roughly |azimuth| ≤ 40°, x > 5 m), matching the
+  standard KITTI 3D detection evaluation region. They are not full 360° sweeps.
+- Attacks target the **Car** class.
+- Intensity is sampled from the *diffuse* range-conditioned distribution rather than
+  saturated. Real spoofers produce saturated returns; forcing saturation makes
+  phantoms trivially detectable as out-of-distribution, so the diffuse choice is
+  what isolates *geometric* robustness. See the paper for the ablation.
+
+## License
+
+`cc-by-nc-sa-3.0`, inherited from KITTI. These are derived point clouds, so the
+non-commercial and share-alike terms of the original apply. Please cite KITTI
+alongside this dataset.
+
+## Citation
+
+This dataset accompanies a paper currently **under review**. It is released
+anonymously for the review period; the citation below will be updated once the
+submission is de-anonymized.
+
+```bibtex
+@inproceedings{atlas_anonymous,
+  title  = {ATLAS: Adversarial LiDAR Attack Splits for 3D Object Detection},
+  author = {Anonymous},
+  note   = {Under review. Dataset released anonymously for peer review.},
+  year   = {2026}
+}
+@inproceedings{geiger2012kitti,
+  title     = {Are we ready for Autonomous Driving? The KITTI Vision Benchmark Suite},
+  author    = {Geiger, Andreas and Lenz, Philip and Urtasun, Raquel},
+  booktitle = {CVPR},
+  year      = {2012}
+}
+@inproceedings{cao2023removal,
+  title     = {You Can't See Me: Physical Removal Attacks on LiDAR-based Autonomous Vehicles Driving Frameworks},
+  author    = {Cao, Yulong and Bhupathiraju, S. Hrushikesh and Naghavi, Pirouz and Sugawara, Takeshi and Mao, Z. Morley and Rampazzi, Sara},
+  booktitle = {USENIX Security},
+  year      = {2023}
+}
+```

baseline_results.csv

@@ -0,0 +1,82 @@
+model,config,split,metric_type,AP_mod,AP_easy_hard,ASR,num_targets,num_success,notes
+pointpillar,cfgs/kitti_models/pointpillar.yaml,clean,detection,76.3261,87.0866|73.2483,,,,Car_3d_R40
+pointpillar,cfgs/kitti_models/pointpillar.yaml,inject_easy,asr,,,0.4243,3769,1599,inject
+pointpillar,cfgs/kitti_models/pointpillar.yaml,inject_medium,asr,,,0.2594,3767,977,inject
+pointpillar,cfgs/kitti_models/pointpillar.yaml,inject_hard,asr,,,0.0482,3649,176,inject
+pointpillar,cfgs/kitti_models/pointpillar.yaml,removal_az10,asr,,,0.3499,3384,1184,removal
+pointpillar,cfgs/kitti_models/pointpillar.yaml,removal_az20,asr,,,0.6637,3384,2246,removal
+pointpillar,cfgs/kitti_models/pointpillar.yaml,removal_az30,asr,,,0.5957,3384,2016,removal
+pointpillar,cfgs/kitti_models/pointpillar.yaml,removal_az40,asr,,,0.5242,3384,1774,removal
+pointpillar,cfgs/kitti_models/pointpillar.yaml,removal_az50,asr,,,0.1974,3384,668,removal
+pointpillar,cfgs/kitti_models/pointpillar.yaml,removal_az60,asr,,,0.1971,3384,667,removal
+second,cfgs/kitti_models/second.yaml,clean,detection,79.3382,88.6033|76.2602,,,,Car_3d_R40
+second,cfgs/kitti_models/second.yaml,inject_easy,asr,,,0.7028,3769,2649,inject
+second,cfgs/kitti_models/second.yaml,inject_medium,asr,,,0.4959,3767,1868,inject
+second,cfgs/kitti_models/second.yaml,inject_hard,asr,,,0.1677,3649,612,inject
+second,cfgs/kitti_models/second.yaml,removal_az10,asr,,,0.2612,3384,884,removal
+second,cfgs/kitti_models/second.yaml,removal_az20,asr,,,0.6534,3384,2211,removal
+second,cfgs/kitti_models/second.yaml,removal_az30,asr,,,0.4329,3384,1465,removal
+second,cfgs/kitti_models/second.yaml,removal_az40,asr,,,0.3209,3384,1086,removal
+second,cfgs/kitti_models/second.yaml,removal_az50,asr,,,0.1330,3384,450,removal
+second,cfgs/kitti_models/second.yaml,removal_az60,asr,,,0.1324,3384,448,removal
+second_iou,cfgs/kitti_models/second_iou.yaml,clean,detection,79.8199,89.2087|77.0521,,,,Car_3d_R40
+second_iou,cfgs/kitti_models/second_iou.yaml,inject_easy,asr,,,0.6938,3769,2615,inject
+second_iou,cfgs/kitti_models/second_iou.yaml,inject_medium,asr,,,0.4630,3767,1744,inject
+second_iou,cfgs/kitti_models/second_iou.yaml,inject_hard,asr,,,0.1477,3649,539,inject
+second_iou,cfgs/kitti_models/second_iou.yaml,removal_az10,asr,,,0.2503,3384,847,removal
+second_iou,cfgs/kitti_models/second_iou.yaml,removal_az20,asr,,,0.6522,3384,2207,removal
+second_iou,cfgs/kitti_models/second_iou.yaml,removal_az30,asr,,,0.4223,3384,1429,removal
+second_iou,cfgs/kitti_models/second_iou.yaml,removal_az40,asr,,,0.3138,3384,1062,removal
+second_iou,cfgs/kitti_models/second_iou.yaml,removal_az50,asr,,,0.1235,3384,418,removal
+second_iou,cfgs/kitti_models/second_iou.yaml,removal_az60,asr,,,0.1215,3384,411,removal
+PartA2,cfgs/kitti_models/PartA2.yaml,clean,detection,81.9240,90.8807|79.7552,,,,Car_3d_R40
+PartA2,cfgs/kitti_models/PartA2.yaml,inject_easy,asr,,,0.4492,3769,1693,inject
+PartA2,cfgs/kitti_models/PartA2.yaml,inject_medium,asr,,,0.1914,3767,721,inject
+PartA2,cfgs/kitti_models/PartA2.yaml,inject_hard,asr,,,0.0310,3649,113,inject
+PartA2,cfgs/kitti_models/PartA2.yaml,removal_az10,asr,,,0.3103,3384,1050,removal
+PartA2,cfgs/kitti_models/PartA2.yaml,removal_az20,asr,,,0.6841,3384,2315,removal
+PartA2,cfgs/kitti_models/PartA2.yaml,removal_az30,asr,,,0.5671,3384,1919,removal
+PartA2,cfgs/kitti_models/PartA2.yaml,removal_az40,asr,,,0.4613,3384,1561,removal
+PartA2,cfgs/kitti_models/PartA2.yaml,removal_az50,asr,,,0.1758,3384,595,removal
+PartA2,cfgs/kitti_models/PartA2.yaml,removal_az60,asr,,,0.1749,3384,592,removal
+PartA2_free,cfgs/kitti_models/PartA2_free.yaml,clean,error,,,,,,missing PartA2_free.pth
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,clean,detection,82.3491,89.5941|80.0481,,,,Car_3d_R40
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,inject_easy,asr,,,0.6676,3769,2516,inject
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,inject_medium,asr,,,0.2947,3767,1110,inject
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,inject_hard,asr,,,0.0485,3649,177,inject
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,removal_az10,asr,,,0.3221,3384,1090,removal
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,removal_az20,asr,,,0.6826,3384,2310,removal
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,removal_az30,asr,,,0.6197,3384,2097,removal
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,removal_az40,asr,,,0.5301,3384,1794,removal
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,removal_az50,asr,,,0.1941,3384,657,removal
+pv_rcnn,cfgs/kitti_models/pv_rcnn.yaml,removal_az60,asr,,,0.1912,3384,647,removal
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,clean,detection,82.5402,89.4378|79.9626,,,,Car_3d_R40
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,inject_easy,asr,,,0.4579,3769,1726,inject
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,inject_medium,asr,,,0.2379,3767,896,inject
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,inject_hard,asr,,,0.0321,3649,117,inject
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,removal_az10,asr,,,0.3579,3384,1211,removal
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,removal_az20,asr,,,0.6817,3384,2307,removal
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,removal_az30,asr,,,0.6188,3384,2094,removal
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,removal_az40,asr,,,0.5160,3384,1746,removal
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,removal_az50,asr,,,0.1894,3384,641,removal
+voxel_rcnn_car,cfgs/kitti_models/voxel_rcnn_car.yaml,removal_az60,asr,,,0.1936,3384,655,removal
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,clean,detection,79.1307,88.6077|75.2059,,,,Car_3d_R40
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,inject_easy,asr,,,0.6423,3769,2421,inject
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,inject_medium,asr,,,0.3154,3767,1188,inject
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,inject_hard,asr,,,0.0422,3649,154,inject
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,removal_az10,asr,,,0.3768,3384,1275,removal
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,removal_az20,asr,,,0.6862,3384,2322,removal
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,removal_az30,asr,,,0.6126,3384,2073,removal
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,removal_az40,asr,,,0.4970,3384,1682,removal
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,removal_az50,asr,,,0.2476,3384,838,removal
+pointrcnn,cfgs/kitti_models/pointrcnn.yaml,removal_az60,asr,,,0.3215,3384,1088,removal
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,clean,detection,79.6597,89.3376|77.5670,,,,Car_3d_R40
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,inject_easy,asr,,,0.6816,3769,2569,inject
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,inject_medium,asr,,,0.3748,3767,1412,inject
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,inject_hard,asr,,,0.0625,3649,228,inject
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,removal_az10,asr,,,0.3478,3384,1177,removal
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,removal_az20,asr,,,0.6865,3384,2323,removal
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,removal_az30,asr,,,0.5804,3384,1964,removal
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,removal_az40,asr,,,0.4498,3384,1522,removal
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,removal_az50,asr,,,0.2196,3384,743,removal
+pointrcnn_iou,cfgs/kitti_models/pointrcnn_iou.yaml,removal_az60,asr,,,0.2982,3384,1009,removal

data/inject_easy/train-00001-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e1d0798feb98736dcb763a4776f9cc4b0c97c3d1447df233625a52787f8675a
+size 118308447

data/inject_easy/train-00004-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f479cb8a8c48465e7e7a5e4857cd4fc902a6ec4fc3279fe57cf868f166a28ff
+size 116633276

data/inject_easy/train-00005-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26b4f0506975d81d298b87e3ea296799eb8d6ba9fb1ca5615c497ab2e3291b84
+size 117729536

data/inject_easy/train-00006-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44f69f030bb0ca7d48f45ae7b88d04adb1e3ce4f731e6b160ad30d09a9bb5c1e
+size 119331170

data/inject_easy/train-00007-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:301cb414139728bbb862ab9fab9ba1e9bb5eb6040beaf4595abf6d7f3dc602a3
+size 61404293

data/inject_hard/train-00000-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4a0cfc4bd8d3e0c9f4c97ffc04d5216e523fa669054fd45c2e5653bd507ac0c
+size 76125730

data/inject_hard/train-00001-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa7ee3f5863b41a127addc16470a6058c66fe7b86396424821415a7876779005
+size 76522693

data/inject_hard/train-00002-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a965c8d172fac1dd8f16fcc78ab230644cbec0eb225a78cec19c9ac910458ce
+size 76378555

data/inject_hard/train-00003-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e52b5fbd27dabb2ffe10629c0cb5db628bf7f64aab0eef2e0b4647c8823c1503
+size 76374725

data/inject_hard/train-00004-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69cd1b777b514133852951935b7de421e3d5ed72985274951eac869d139af101
+size 76566727

data/inject_hard/train-00005-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cec0fccb9774d02d4ee02ea320126ffa9f9c9ab0e1f8b68dd7dde37fb453a7c
+size 76426585

data/inject_hard/train-00006-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c2c0a6c1cd5dacee16d5bbd0750fdd0073ffd70c360dca6139739b57e987fa7
+size 75743682

data/inject_hard/train-00007-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:900e97641afc8786fe0801ce81cd587509a0720e4b66abc1ba8c51da3771bf97
+size 22787000

data/removal_az20/train-00000-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0abde536c0eff98eff9eb04d832ed3d6c2099feb21b7dc2ed9ab8cda846d9a88
+size 66203659

data/removal_az20/train-00001-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cd4d8f1499acb1b844f1805e89906a9b5d5eb268852657a821174e916b8ad78
+size 65332058

data/removal_az20/train-00002-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:beee6a270f28ab62a8c6b8d6a0ecb9c19ff6942439d94b77cea01a86c82babfa
+size 66408693

data/removal_az20/train-00003-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2bc547d2c71b448448cce768a8966db87f61682b536b6b57e7f01859d93e4f8
+size 66308706

data/removal_az20/train-00004-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1df29842a413bb39aba3409d321ad523064ea531c232d4d18c5eba2c948f9ad
+size 64898122

data/removal_az20/train-00005-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bfc014da786713d8e6a4f1efec8c8b64f303e2a5e6411fe91b1aed29d4b449a
+size 66487808

data/removal_az20/train-00006-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd56e4d83956bfb883776c9ae0285690f7bbefca405cb1ecbe7caed6bc1c15c4
+size 50587008

data/removal_az30/train-00000-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e21eaaa6c9bf7744c5460e843fa48e513645ac0eab0f02b62a2b5bde42a1196c
+size 62173120

data/removal_az30/train-00001-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3a85b73736db7737e1b6ab082c89e391514e85d55eaed7e8c0cbd6d925d11a4
+size 60885360

data/removal_az30/train-00002-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b2f5f536973248b16140600964b9c3932ddc6a3312573e14d0ee9c486294173
+size 62130086

data/removal_az30/train-00003-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a320877790014f93f094a8abece5e283b65da76378e77d1577dbb3372fefe26f
+size 61852626

data/removal_az30/train-00004-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:686022adc491d70330d5c234fe206ade3813b5b2ca82aab18252fdc4d27f0a6a
+size 60452197

data/removal_az30/train-00005-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1fbecc92745caa16f93201095236642ea36557cbe34a0f1642711ecba83882f0
+size 62091560

data/removal_az30/train-00006-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d79438d2d7de2aa8c9d551009288151b1e5207dfbb919085125c703cf7fb384c
+size 47221487

data/removal_az40/train-00000-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f009ab4d9326f2cb8f00ddbe0b4d1d703dec3f8da68274b0a2ffb896cdf147ea
+size 58180805

data/removal_az40/train-00001-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:004ce24fa2376ba09d77a9cb9ea2f7141f7647924aa7744aaa03f68d6c000f01
+size 56633413

data/removal_az40/train-00002-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2690f59985be27a5dc21ed643e39bab5c1e57e3cc85e4f5417fbcf027e10c13
+size 57927330

data/removal_az40/train-00003-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e017819eb5ef9c53052d70531562ca6e41e71b2fa2fdb398409ab8fec1cd9afa
+size 57624101

data/removal_az40/train-00004-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84dce4f8896a1749f6a55b40b66f09d407af6215a49e45dec0b027eb6dbd14e3
+size 55927502

data/removal_az40/train-00005-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d05c6d8ce7ba827b0348412a23e56ede9be32ee173602d2b5ac73e6f36f1bab
+size 58141464

data/removal_az40/train-00006-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54ca21e7ff3e7690f3b1e2bbb442032783de867409727cbeed705c6c73f27be2
+size 44399684

data/removal_az50/train-00000-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b5f996461a7f4c4d4ffd51fa524ef9ecacd2e1cbe14ebd3bbf2a3bb632042e9
+size 56060321

data/removal_az50/train-00001-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c15aa796f82542e3e83d084daff2e4e791d30cde0bcac2f05f96f716394bf2a1
+size 54514062

data/removal_az50/train-00002-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8da44d52a808a184fc74a8cd9c923989d6af8a5750a49582937848c494a3b1db
+size 55567030

data/removal_az50/train-00003-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bed35a883d958f676c11ad9847c028905b176300345514fa01a2451e75d76604
+size 55326627

data/removal_az50/train-00004-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cc8f049cf1322e26c83b8ad459dd144846ece83a32d0e062bb94cad2c47bc51
+size 53772134

data/removal_az50/train-00005-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b1a7081ed61433576f19455b03241bb3f338dbe807c788c63e09d04592b11c7
+size 55954966

data/removal_az50/train-00006-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b5309fc1ea0889b819591544ca8e7ca43d17ae097bba0ea76b260747f89d7b9d
+size 42695330

data/removal_az60/train-00000-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:685401637d7960c182e97c6ce5b96dbb36492410f9db0cbc09b9af85546b0873
+size 52794378

data/removal_az60/train-00001-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2afbacca9c33f13e7ac594204c70203a3574b908fe05759bf161b2b054139b4
+size 50962563

data/removal_az60/train-00004-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:775f50a6d71729c47a6701a237b6b0c2c1a5d66463849ab036db44b2a99d95d3
+size 50127195

data/removal_az60/train-00005-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:100f75957324134429b8ffb1a235b035e21958eabfe5fa30375e129d4b74c3a7
+size 52595702

data/removal_az60/train-00006-of-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:939a7e0dc1a3bab22eec80462933cd518d6e183a61a77479d60f344bf7524e87
+size 40236977

load_atlas_kitti.py

@@ -0,0 +1,98 @@
+#!/usr/bin/env python
+"""Convenience loader for ATLAS-KITTI.
+
+Point clouds are stored FLAT in Parquet (a flat float32 array plus `num_points`
+and `point_dim`) because Parquet handles a flat array far more efficiently than a
+nested list-of-lists. These helpers hide that detail, so you get `(N, 4)` arrays
+directly.
+
+    from load_atlas_kitti import load_atlas, to_arrays
+
+    ds = load_atlas("removal_az20")
+    for ex in ds:
+        pts, spoof_gt, gt_boxes, gt_names = to_arrays(ex)
+        # pts: (N, 4) float32  = x, y, z, intensity
+"""
+import numpy as np
+
+REPO = "ps3020/atlas-kitti"
+
+INJECT_CONFIGS = ["inject_easy", "inject_medium", "inject_hard"]
+REMOVAL_CONFIGS = [f"removal_az{a}" for a in (10, 20, 30, 40, 50, 60)]
+CONFIGS = INJECT_CONFIGS + REMOVAL_CONFIGS
+
+
+def load_atlas(config, split="train", repo=REPO, **kw):
+    """load_dataset wrapper. Pass streaming=True to avoid materialising the split."""
+    from datasets import load_dataset
+    if config not in CONFIGS:
+        raise ValueError(f"unknown config {config!r}; expected one of {CONFIGS}")
+    return load_dataset(repo, config, split=split, **kw)
+
+
+def points_of(example):
+    """The attacked cloud as (N, 4) float32: x, y, z, intensity."""
+    return np.asarray(example["points"], dtype=np.float32).reshape(
+        example["num_points"], example["point_dim"])
+
+
+def gt_of(example):
+    """Clean KITTI ground truth as ((M, 7) boxes, [M] class names)."""
+    n = example["num_gt"]
+    boxes = np.asarray(example["gt_boxes"], dtype=np.float32).reshape(n, 7)
+    return boxes, list(example["gt_names"])
+
+
+def to_arrays(example):
+    """(points (N,4), spoof_gt (7,), gt_boxes (M,7), gt_names [M])."""
+    boxes, names = gt_of(example)
+    return (points_of(example),
+            np.asarray(example["spoof_gt"], dtype=np.float32),
+            boxes, names)
+
+
+def attack_type(config):
+    return "inject" if config.startswith("inject") else "removal"
+
+
+def score_frame(example, config, detected_overlapping_spoof_gt):
+    """Whether the attack SUCCEEDED on this frame.
+
+    injection: success = a detection overlaps the phantom (a false positive was
+               created).
+    removal  : success = NO detection overlaps the erased object (a true positive
+               was destroyed).
+
+    `detected_overlapping_spoof_gt` is your detector's answer to "is there a
+    predicted box with IoU >= threshold against example['spoof_gt']?" -- ATLAS
+    reports ASR at IoU 0.3.
+    """
+    hit = bool(detected_overlapping_spoof_gt)
+    return hit if attack_type(config) == "inject" else (not hit)
+
+
+def asr(successes, dataset_or_len):
+    """Attack success rate.
+
+    IMPORTANT: the denominator is the number of frames in THIS config, not 3769.
+    Frames that could not be attacked were never written, so configs have
+    different lengths (3769 / 3767 / 3649 for injection, 3384 for removal).
+    Dividing by 3769 understates removal ASR by about 10%.
+    """
+    n = dataset_or_len if isinstance(dataset_or_len, int) else len(dataset_or_len)
+    return (successes / n) if n else float("nan")
+
+
+def removal_rate(example):
+    """Realised fraction of in-wedge points deleted (removal configs only)."""
+    n_in = example.get("atk_n_points_in_sector") or 0
+    return (example["atk_n_points_removed"] / n_in) if n_in else float("nan")
+
+
+if __name__ == "__main__":
+    ds = load_atlas("removal_az20", streaming=True)
+    ex = next(iter(ds))
+    pts, sg, boxes, names = to_arrays(ex)
+    print(f"frame {ex['frame_id']}: points {pts.shape}, spoof_gt {sg.shape}, "
+          f"{len(names)} GT objects {names}")
+    print(f"realised removal rate: {removal_rate(ex):.3f}")