Rename benchmark release layout to benchmark_splits and data

.gitattributes

@@ -68,3 +68,8 @@ benchmark_data/test_expert_unseen/video_progress_benchmark_file.json filter=lfs
 benchmark_data/test_nonexpert_seen/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
 benchmark_data/test_nonexpert_unseen/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
 benchmark_data/train/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
+benchmark_splits/train/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
+benchmark_splits/test_expert_seen/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
+benchmark_splits/test_expert_unseen/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
+benchmark_splits/test_nonexpert_seen/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text
+benchmark_splits/test_nonexpert_unseen/video_progress_benchmark_file.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -14,16 +14,16 @@ VLAC2 is a video-progress benchmark for long-horizon robot manipulation. This re
 
 ## Contents
 
-- `benchmark_data/train/video_progress_benchmark_file.json`
-- `benchmark_data/test_expert_seen/video_progress_benchmark_file.json`
-- `benchmark_data/test_expert_unseen/video_progress_benchmark_file.json`
-- `benchmark_data/test_nonexpert_seen/video_progress_benchmark_file.json`
-- `benchmark_data/test_nonexpert_unseen/video_progress_benchmark_file.json`
+- `benchmark_splits/train/video_progress_benchmark_file.json`
+- `benchmark_splits/test_expert_seen/video_progress_benchmark_file.json`
+- `benchmark_splits/test_expert_unseen/video_progress_benchmark_file.json`
+- `benchmark_splits/test_nonexpert_seen/video_progress_benchmark_file.json`
+- `benchmark_splits/test_nonexpert_unseen/video_progress_benchmark_file.json`
 - `scripts/extract_vlac2_release_raw_archives.sh`
 - `scripts/extract_vlac2_release_frames.py`
 - `scripts/vlac2_release_common.py`
-- `video_archives/vlac2_benchmark_train_main_path_videos.tar`
-- `video_archives/vlac2_benchmark_test_main_path_videos.tar`
+- `data/train_videos.tar`
+- `data/test_videos.tar`
 
 ## Usage
 
@@ -31,11 +31,12 @@ VLAC2 is a video-progress benchmark for long-horizon robot manipulation. This re
 
 ```bash
 bash scripts/extract_vlac2_release_raw_archives.sh \
-  /path/to/vlac2_release/video_archives \
+  /path/to/vlac2_release/data \
   /path/to/vlac2_release/vlac2_release_data
 ```
 
-If both arguments are omitted, the script defaults to `video_archives/` and extracts into `vlac2_release_data/` under the release root.
+If both arguments are omitted, the script defaults to `data/` and extracts into `vlac2_release_data/` under the release root.
+The script expects exactly the released benchmark archives: `test_videos.tar` and `train_videos.tar` (or their `.tar.zst` variants).
 
 2. Extract benchmark frames from the unpacked raw videos.
 

scripts/extract_vlac2_release_frames.py

@@ -53,7 +53,7 @@ def parse_args() -> argparse.Namespace:
         "--release-root",
         type=Path,
         required=True,
-        help="Release root containing benchmark_data/ and vlac2_release_data/.",
+        help="Release root containing benchmark_splits/ and vlac2_release_data/.",
     )
     parser.add_argument(
         "--raw-root",
@@ -71,7 +71,7 @@ def parse_args() -> argparse.Namespace:
         "--benchmark-json",
         action="append",
         default=None,
-        help="Optional specific benchmark JSON(s). Defaults to discovering benchmark_data/**/*.json inside the release root.",
+        help="Optional specific benchmark JSON(s). Defaults to discovering benchmark_splits/**/*.json inside the release root.",
     )
     parser.add_argument(
         "--view-mode",

scripts/extract_vlac2_release_raw_archives.sh

@@ -2,27 +2,35 @@
 set -euo pipefail
 
 ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-ARCHIVE_DIR="${1:-${ROOT_DIR}/video_archives}"
+ARCHIVE_DIR="${1:-${ROOT_DIR}/data}"
 RAW_DIR="${2:-${ROOT_DIR}/vlac2_release_data}"
+EXPECTED_BASES=(
+  "test_videos"
+  "train_videos"
+)
 
 mkdir -p "${RAW_DIR}"
 
-mapfile -d '' -t archives < <(
-  find "${ARCHIVE_DIR}" -maxdepth 1 -type f \
-    \( -name '*.tar' -o -name '*.tar.zst' \) \
-    -print0 | LC_ALL=C sort -z
-)
-
-if [[ "${#archives[@]}" -eq 0 ]]; then
-  echo "no archives found under ${ARCHIVE_DIR}" >&2
-  exit 1
-fi
+archives=()
+for base in "${EXPECTED_BASES[@]}"; do
+  matches=()
+  for suffix in ".tar" ".tar.zst"; do
+    candidate="${ARCHIVE_DIR}/${base}${suffix}"
+    if [[ -f "${candidate}" ]]; then
+      matches+=("${candidate}")
+    fi
+  done
 
-for archive in "${archives[@]}"; do
-  if [[ ! -f "${archive}" ]]; then
-    echo "missing archive: ${archive}" >&2
+  if [[ "${#matches[@]}" -eq 0 ]]; then
+    echo "missing expected archive under ${ARCHIVE_DIR}: ${base}.tar or ${base}.tar.zst" >&2
     exit 1
   fi
+  if [[ "${#matches[@]}" -gt 1 ]]; then
+    echo "found multiple archive variants for ${base} under ${ARCHIVE_DIR}; keep only one of .tar or .tar.zst" >&2
+    exit 1
+  fi
+
+  archives+=("${matches[0]}")
 done
 
 for archive in "${archives[@]}"; do

scripts/vlac2_release_common.py

@@ -24,7 +24,7 @@ GENERATED_MARKER_FILENAME = "_GENERATED"
 PUBLIC_BENCHMARK_JSON_NAME = "video_progress_benchmark_file.json"
 PUBLIC_RELEASE_ROOT_PLACEHOLDER = "__VLAC2_RELEASE_ROOT__"
 PUBLIC_FRAMES_ROOT_PLACEHOLDER = "__VLAC2_FRAMES_ROOT__"
-PUBLIC_BENCHMARK_DIRNAME = "benchmark_data"
+PUBLIC_BENCHMARK_DIRNAME = "benchmark_splits"
 LEGACY_BENCHMARK_DIRNAME = "benchmark_json"