Refine README style for Video-Progress Benchmark release

README.md

@@ -10,56 +10,118 @@ license: other
 
 # Video-Progress Benchmark Release
 
-Video-Progress Benchmark is a benchmark for long-horizon robot manipulation progress prediction. This release provides the benchmark JSON files, the minimal raw-video subset required by the released protocol, and a lightweight frame-extraction workflow for reproducible evaluation.
+Video-Progress Benchmark is a benchmark for long-horizon robot manipulation progress prediction. This release provides benchmark JSON files, the minimal raw-video subset required by the released protocol, and a lightweight frame-extraction workflow for reproducible evaluation.
 
 ## Contents
 
-- `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`
-- `data/train_videos.tar`
-- `data/test_videos.tar`
+```text
+benchmark_splits/
+  train/video_progress_benchmark_file.json
+  test_expert_seen/video_progress_benchmark_file.json
+  test_expert_unseen/video_progress_benchmark_file.json
+  test_nonexpert_seen/video_progress_benchmark_file.json
+  test_nonexpert_unseen/video_progress_benchmark_file.json
+
+scripts/
+  extract_vlac2_release_raw_archives.sh
+  extract_vlac2_release_frames.py
+
+data/
+  train_videos.tar
+  test_videos.tar
+```
 
 ## Usage
 
-1. Unpack the video archives.
+### 1. Unpack the video archives
 
 ```bash
 bash scripts/extract_vlac2_release_raw_archives.sh
 ```
 
-This extracts `data/test_videos.tar` and `data/train_videos.tar` into `data/`.
+This extracts `data/train_videos.tar` and `data/test_videos.tar` into `data/`.
+
 To extract the raw videos elsewhere, pass a target directory:
 
 ```bash
 bash scripts/extract_vlac2_release_raw_archives.sh /path/to/data
 ```
 
-2. Extract frames.
+### 2. Extract frames
 
 ```bash
 python scripts/extract_vlac2_release_frames.py --data-root data
 ```
 
-This writes extracted frames to `_extracted_frames/` under the release directory.
-Only the benchmark main view is extracted.
+By default, extracted frames are written to `_extracted_frames/` under the release directory. Only the benchmark main view is extracted.
+
 If the raw videos were extracted elsewhere:
 
 ```bash
 python scripts/extract_vlac2_release_frames.py --data-root /path/to/data
 ```
 
-3. Use the benchmark splits.
+### 3. Use the benchmark splits
+
+The benchmark JSON files are ready to use. No benchmark reconstruction is required.
+
+Frame paths in the JSON files use the prefix:
+
+```text
+__VLAC2_FRAMES_ROOT__/
+```
+
+Replace `__VLAC2_FRAMES_ROOT__` with the absolute path to the extracted-frame directory.
 
-Frame paths in the JSON files use the prefix `__VLAC2_FRAMES_ROOT__/...`.
-Replace `__VLAC2_FRAMES_ROOT__` with the absolute path of `_extracted_frames`.
+For example, if frames are extracted to:
+
+```text
+/path/to/release/_extracted_frames
+```
+
+then:
+
+```text
+__VLAC2_FRAMES_ROOT__/...
+```
+
+should be resolved as:
+
+```text
+/path/to/release/_extracted_frames/...
+```
+
+## Benchmark Splits
+
+The release contains one training split and four evaluation splits:
+
+* `train`
+* `test_expert_seen`
+* `test_expert_unseen`
+* `test_nonexpert_seen`
+* `test_nonexpert_unseen`
+
+The test splits are intended for evaluation only.
+
+## Dependencies
+
+A video decoder backend is required for frame extraction. Install one of:
+
+```bash
+pip install opencv-python
+```
+
+```bash
+pip install av
+```
+
+```bash
+pip install imageio imageio-ffmpeg
+```
 
 ## Notes
 
-- A video decoder backend is required for frame extraction: `opencv-python`, `av` (PyAV), or `imageio`.
-- This repo only contains the raw videos required by the released benchmark splits. The full raw-data release is distributed separately.
-- The benchmark JSON files are ready to use. No benchmark reconstruction is required.
+* This repository only contains the raw videos required by the released benchmark splits.
+* The full raw-data release is distributed separately.
+* The benchmark JSON files are pre-generated and ready to use.
+* The raw-video archives and extracted-frame cache are separated so that users can regenerate frames under their own storage layout.