add README.md

README.md

@@ -0,0 +1,40 @@
+---
+license: other
+task_categories:
+- image-to-image
+tags:
+- novel-view-synthesis
+- autonomous-driving
+- lidar
+- multi-view
+size_categories:
+- 10K<n<100K
+---
+
+# MHDCSM/cv_ds
+
+WebDataset-format shards of the autonomous driving novel-view-synthesis dataset.
+
+Splits:
+- `train-*.tar` — training samples (includes ground-truth target images)
+- `test-*.tar`  — test samples (no targets)
+
+Each tar holds many samples. Per-sample files share a key (the `sample_id`):
+
+| Suffix | Content |
+|--------|---------|
+| `meta.json` | camera poses, intrinsics, timestamps, target camera name |
+| `t0.<cam>.jpg`, `t1.<cam>.jpg` | 6 camera views per timestamp |
+| `lidar.npz` | dense world-frame point cloud (xyz, intensity) |
+| `target.<cam>.jpg` | ground-truth target image (train only) |
+
+## Streaming usage
+
+```python
+import webdataset as wds
+url = "https://huggingface.co/datasets/MHDCSM/cv_ds/resolve/main/train-{000000..NNNNNN}.tar"
+ds = wds.WebDataset(url).decode()
+for sample in ds:
+    meta = sample["meta.json"]
+    img  = sample["t0.front.jpg"]
+```