Updated examples and script organization

Makefile

@@ -0,0 +1,15 @@
+.PHONY: prepare webdataset parquet clean
+
+prepare:
+	mkdir -p downloads
+	python scripts/prepare.py downloads data manifest.csv
+
+webdataset:
+	mkdir -p shards
+	python scripts/build_webdataset.py data shards
+
+parquet:
+	python scripts/build_parquet.py
+
+clean:
+	rm -vr downloads

README.md

@@ -89,13 +89,13 @@ python -m pip install cityscapesscripts
    Note that this may prompt your [Cityscapes account](https://cityscapes-dataset.com/login/) login credentials.
 
 ```bash
-python csvps.py
+make prepare
 ```
 
 3. Remove the downloaded Cityscapes archive files to save disk space (optional).
 
 ```bash
-rm -r downloads
+make clean
 ```
 
 ## Usage
@@ -104,7 +104,7 @@ Due to the structure, the dataset can be easily loaded using the `webdataset` li
 To convert the `train`, `val` and `test` directories into a `tar` archive, run the following command:
 
 ```bash
-for split in train val test; do tar -cf "shards/csvps-$split-000000.tar" "data/$split" done
+make webdataset
 ```
 
 Subsequently, the dataset can be loaded using the `webdataset` library:
@@ -112,7 +112,7 @@ Subsequently, the dataset can be loaded using the `webdataset` library:
 ```python
 import webdataset as wds
 
-dataset = wds.WebDataset("shards/csvps-train-{000000}.tar")
+dataset = wds.WebDataset("shards/train.tar")
 ```
 
 

examples.ipynb

@@ -0,0 +1,78 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## WebDataset\n",
+    "\n",
+    "The dataset can be exported to a `.tar` archive and iterated with the `webdataset`\n",
+    "package.\n",
+    "\n",
+    "After building the WebDataset-formatted archives using `make webdataset`, the dataset can be iterated as follows."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 23,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "000000/000000: image.png timestamp.txt vehicle.json \n",
+      "000000/000001: image.png timestamp.txt vehicle.json \n",
+      "000000/000002: image.png timestamp.txt vehicle.json \n",
+      "000000/000003: image.png timestamp.txt vehicle.json \n",
+      "000000/000004: depth.tiff image.png panoptic.png timestamp.txt vehicle.json \n",
+      "000000/000005: image.png timestamp.txt vehicle.json \n",
+      "000000/000006: image.png timestamp.txt vehicle.json \n",
+      "000000/000007: image.png timestamp.txt vehicle.json \n",
+      "000000/000008: image.png timestamp.txt vehicle.json \n",
+      "000000/000009: depth.tiff image.png panoptic.png timestamp.txt vehicle.json \n",
+      "000000/000010: image.png timestamp.txt vehicle.json \n"
+     ]
+    }
+   ],
+   "source": [
+    "import webdataset as wds\n",
+    "\n",
+    "# Create iterable dataset\n",
+    "ds = wds.WebDataset(\"shards/csvps-val.tar\", shardshuffle=False, verbose=True)\n",
+    "\n",
+    "# Iterate over the dataset and print the keys and the first few samples\n",
+    "for i, sample in enumerate(ds):\n",
+    "    if i > 10:  \n",
+    "        break\n",
+    "    print(sample[\"__key__\"], end=\": \")\n",
+    "    for k in sample:\n",
+    "        if not k.startswith(\"__\"):\n",
+    "            print(k, end=\" \")\n",
+    "    print(flush=True)\n",
+    "    "
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

requirements.txt

@@ -0,0 +1,2 @@
+cityscapesscripts
+webdataset

scripts/build_parquet.py

@@ -0,0 +1,2 @@
+print("Building a Parquet dataset is not (yet) implemented. Use WebDataset instead.")
+exit(1)

scripts/build_webdataset.py

@@ -0,0 +1,49 @@
+#!/usr/bin/env python
+r"""
+Builds a WebDataset from the Cityscapes Video dataset.
+"""
+
+import argparse
+import tarfile
+
+from pathlib import Path
+from tqdm import tqdm
+
+
+def parse_args():
+    p = argparse.ArgumentParser(
+        description="Build a WebDataset from the Cityscapes Video dataset."
+    )
+    p.add_argument("--prefix", default="csvps", help="Prefix for the tar files.")
+    p.add_argument("data", type=Path, help="Path to the Cityscapes Video dataset.")
+    p.add_argument("output", type=Path, help="Path to the output directory.")
+
+    return p.parse_args()
+
+
+def build_dataset(split: str, data_dir: Path, out_dir: Path, *, prefix: str = ""):
+    data_dir = data_dir / split
+    name = f"{split}.tar"
+    if prefix and prefix != "":
+        name = f"{prefix}-{name}"
+    tar_path = out_dir / name
+
+    if tar_path.exists():
+        print(f"Error: Tar archive already exists: {tar_path}")
+
+    with tarfile.open(tar_path, "w") as tar:
+        # Add the files to the tar archive
+        for file in tqdm(sorted(data_dir.glob("**/*")), desc=f"Building {tar_path}"):
+            tar.add(file, arcname=file.relative_to(data_dir))
+
+
+def main():
+    args = parse_args()
+
+    # For each split, build a tar archive containing the sorted files
+    for split in ("train", "val", "test"):
+        build_dataset(split, args.data, args.output, prefix=args.prefix)
+
+
+if __name__ == "__main__":
+    main()

csvps.py → scripts/prepare.py

@@ -32,19 +32,16 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "downloads_dir",
         type=Path,
-        default=Path("downloads"),
         help="Path to the directory where ZIP files are/will be stored (e.g., 'downloads').",
     )
     parser.add_argument(
         "data_dir",
         type=Path,
-        default=Path("data"),
         help="Path to the directory where extracted files should be moved (e.g., 'data').",
     )
     parser.add_argument(
         "manifest_file",
         type=Path,
-        default=Path("manifest.csv"),
         help="Path to the manifest file (e.g., 'manifest.csv').",
     )
     return parser.parse_args()