added notebook. Fixed 4 npz files that were broken. Small updates overall.

.gitignore

@@ -9,4 +9,6 @@ copyfiles3.txt
 raw_data/npz_540p
 exception.txt
 __pycache__
+error.txt
+sanity_check.py
 

README.md

@@ -189,11 +189,19 @@ Basic usage:
  'semantic_segprop8': torch.Size([5, 540, 960])}
 ```
 
-</summary>
+</details>
 
 ## TODOs
+- Fix remaining bad npz files
+```
+/scratch/sdc/datasets/dronescapes/data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3565.npz
+/scratch/sdc/datasets/dronescapes/data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3570.npz
+/scratch/sdc/datasets/dronescapes/data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3582.npz
+/scratch/sdc/datasets/dronescapes/data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3592.npz
+```
+
+- convert camera normals to world normals
 - mask2former convert
 - add raw script for reading data
   - add semantics for each representation in a DronescapesReader
 - add notebook for visualisation
-- push to huggingface dataset

data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3565.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e026953399a94aa35c9d5721c5c0aa0fa3b08a8d1c1497e49c276579e7862ed
-size 4173
+oid sha256:9f0f4de75392b21d1a4dc0df1ea964e8151ccbe929790e1538cb0ae1f96a8fd6
+size 2073864

data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3570.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3560e1180efb998cc14e8a43d587f0cd0bdeb52fbab1790d31d7fa8676df958b
-size 4173
+oid sha256:0ab9ca4ac88fd68f28f36ce7cffaa5b813b6c894d525292a36240cb9e0ff0032
+size 2073864

data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3582.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4b9976f2c862de40fce5007e9b2974a91a1ecb6428b680aac33510034964183
-size 4173
+oid sha256:87797811c59b4ddaa5a404a88f91728b11810d6faccfafc231457af8feef9658
+size 2073864

data/semisupervised_set/depth_dpt/part0/herculane_DJI_0021_full_3592.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df7b94689301b1485ab2e885fe81306b674c9ad847098fb7eed1af2740fe8568
-size 4173
+oid sha256:3c8eb3bc06c9e7dd3e18c3c27b37ab182f866ef295e64119e4faf4a5f3fabe76
+size 2073864

dronescapes_reader/multitask_dataset.py

@@ -38,7 +38,8 @@ class NpzRepresentation:
         if x.shape[-1] != 3: x = x[..., 0:1]
         if x.shape[-1] == 1: x = x.repeat(1, 1, 3)
         x = x.nan_to_num(0).cpu().numpy() # guaranteed to be (H, W, 3) at this point hopefully
-        if x.dtype != np.uint8: x = np.nan_to_num((x - x.min() / (x.max() - x.min())) * 255, 0).astype(np.uint8)
+        _min, _max = x.min((0, 1), keepdims=True), x.max((0, 1), keepdims=True)
+        if x.dtype != np.uint8: x = np.nan_to_num((x - _min) / (_max - _min) * 255, 0).astype(np.uint8)
         return x
 
     def __repr__(self):
@@ -69,7 +70,7 @@ class MultiTaskDataset(Dataset):
 
     def __init__(self, path: Path, task_names: list[str] | None = None, handle_missing_data: str = "fill_none",
                  files_suffix: str = "npz"):
-        assert path.exists(), f"Provided path '{path}' doesn't exist!"
+        assert Path(path).exists(), f"Provided path '{path}' doesn't exist!"
         assert handle_missing_data in ("drop", "fill_none"), f"Invalid handle_missing_data mode: {handle_missing_data}"
         assert files_suffix == "npz", "Only npz supported right now (though trivial to update)"
         self.path = Path(path).absolute()