Simplify HelixBench dataset card

README.md

@@ -7,48 +7,71 @@ configs:
     path: data/test.zip
 ---
 
-# Helixbench
+# HelixBench
 
-A frozen 1,015-sample audio-video benchmark. This public release contains the **first frame** and **`caption_av` only** for each sample, plus the six-metric evaluation entrypoint. The sample IDs and prompts are fixed.
+**A benchmark for audio-video generation and viewpoint-aligned spatial sound.**
 
-| Metric | Scope | Protocol |
+HelixBench contains **1,015 fixed test cases** and evaluates generated videos with six metrics. Each case provides a first frame and one audio-video caption (`caption_av`). See the [HelixWorld paper](https://arxiv.org/abs/2609.38123) for the benchmark motivation and results.
+
+## What's included
+
+| File | Contents |
+|---|---|
+| [`data/test.zip`](https://huggingface.co/datasets/NoizAI/Helixbench/tree/main/data) | 1,015 first-frame images and their metadata |
+| [`metadata.jsonl`](https://huggingface.co/datasets/NoizAI/Helixbench/blob/main/metadata.jsonl) | Sample IDs and AV captions in a directly readable file |
+| [`evaluation.zip`](https://huggingface.co/datasets/NoizAI/Helixbench/blob/main/evaluation.zip) | Evaluation code and frozen reference features |
+
+The release contains **AV captions only**. Original and generated videos are not bundled; provide your model's generated MP4 for each sample ID.
+
+## Metrics
+
+| Metric | Cases | Measures |
 |---|---:|---|
-| KL ↓ | 1,015 | PaSST 527-class softmax KL(GT‖generated), paired by ID |
-| FAD ↓ | 1,015 | VGGish features on the first five seconds, Fréchet distance |
-| DeSync (s) ↓ | 165 event | Synchformer mean absolute offset |
-| IB ↑ | 1,015 | ImageBind paired video/audio cosine |
-| CLAP ↑ | 1,015 | LAION-CLAP generated audio against `caption_av` |
-| Spatial ↑ | 150 perspective | v3 all-150 mean: invalid `-100`, eligible center `0`, other clips use signed lateral-frame correctness |
+| KL ↓ | 1,015 | Audio distribution similarity using PaSST |
+| FAD ↓ | 1,015 | Audio quality using VGGish features |
+| DeSync (s) ↓ | 165 event cases | Audio-video synchronization |
+| ImageBind ↑ | 1,015 | Video-audio alignment |
+| CLAP ↑ | 1,015 | Generated audio against `caption_av` |
+| Spatial ↑ | 150 perspective cases | Whether spatial sound follows viewpoint motion |
 
-The dataset is `data/test.zip` in ImageFolder format. `metadata.jsonl` also exposes the AV-only captions and IDs as a directly readable file. The evaluation source is browsable under `evaluation/`; `evaluation.zip` additionally bundles the verified GT features. Each row has `sample_id`, `caption_av`, `scene`, `interaction`, and `first_frame_sha256`; `file_name` points to the PNG inside the archive. No video-only or audio-only captions are in the release. `evaluation.zip` contains the Python package and frozen GT PaSST/VGGish features. The Spatial target vocabulary is frozen in `evaluation/src/helixbench/resources/spatial_targets150.jsonl`; this preserves the historical detector prompts without releasing other captions.
+Spatial uses the **v3 all-150 protocol**: invalid clips score −100, eligible center-only clips score 0, and the remaining clips use signed lateral-frame correctness. The six metrics are reported separately; there is no combined score.
 
-## Install the evaluation environment
+## Quick start
 
-On Linux with an NVIDIA GPU, a compatible driver, `python3`, `git`, `curl`, `unzip`, `ffmpeg`, and `ffprobe`, run:
+On Linux with a compatible NVIDIA GPU, install `python3`, `git`, `curl`, `unzip`, `ffmpeg`, and `ffprobe`, then run:
 
 ```bash
 mkdir -p helixbench && cd helixbench
-curl -fL --retry 3 -o evaluation.zip https://huggingface.co/datasets/NoizAI/Helixbench/resolve/main/evaluation.zip
+curl -fL --retry 3 -o evaluation.zip \
+  https://huggingface.co/datasets/NoizAI/Helixbench/resolve/main/evaluation.zip
 unzip -q evaluation.zip
+
 R="$PWD/runtime"
 bash evaluation/install.sh "$R"
 "$R/env/bin/python" evaluation/download_weights.py "$R"
 "$R/env/bin/helixbench" verify-package
 ```
 
-`install.sh` creates a Python 3.12.15 environment with PyTorch 2.7.0/CUDA 12.8, clones four public model repositories at fixed Git commits, applies the included ImageBind compatibility shim, and installs the evaluation package. `download_weights.py` downloads public weights and checks their SHA256, including PaSST and Grounding DINO. Keep the unpacked `evaluation` directory: the editable installation and packaged GT features refer to it. See [the exact dependency list](evaluation/provenance/DEPENDENCIES.md).
+Prepare a `generated.jsonl` containing exactly one MP4 per sample ID. Each row needs `sample_id`, an absolute `path`, and the file's `sha256`. See the [input format](evaluation/examples/INPUT_FORMAT.md) for the media requirements.
 
-## Evaluate generated videos
+```bash
+"$R/env/bin/helixbench" prepare \
+  --generated /absolute/generated.jsonl \
+  --model-id YOUR_MODEL \
+  --run-dir /absolute/run
 
-Provide one MP4 for every sample ID, at 768×512, 121 frames/24 fps, stereo 48 kHz audio with at least 5 seconds. The generated JSONL rows must contain `sample_id`, absolute `path`, and SHA256 `sha256`. A complete manifest has exactly 1,015 rows.
+"$R/env/bin/helixbench" run \
+  --run-dir /absolute/run --metric all \
+  --deps-root "$R/deps" \
+  --weights-root "$R/weights" \
+  --torch-home "$R/torch"
 
-```bash
-"$R/env/bin/helixbench" prepare --generated /absolute/generated.jsonl --model-id my_model --run-dir /absolute/run
-"$R/env/bin/helixbench" run --run-dir /absolute/run --metric all --deps-root "$R/deps" --weights-root "$R/weights" --torch-home "$R/torch"
 "$R/env/bin/helixbench" aggregate --run-dir /absolute/run
 "$R/env/bin/helixbench" verify --run-dir /absolute/run
 ```
 
-The included GT features are verified by SHA256, so GT videos are not needed for KL/FAD. `--gt` remains available if you have the exact GT media and want to extract those features again. The entrypoint produces exactly six metrics; it does not combine them into a single score.
+The installer pins the third-party source revisions and checks downloaded weights. See [dependencies and provenance](evaluation/provenance/DEPENDENCIES.md) for exact versions and hashes. The full workflow has been tested on one NVIDIA A10G.
+
+## Citation
 
-`data/test/metadata.jsonl` and all 1,015 PNG files are SHA-checked against the frozen release manifest before upload. The package's `verify-package` checks the frozen cohort, Spatial vocabulary and scoring engine hashes. Evaluation code is based on the private GitHub Helixbench repository at commit `1f9b199`, with an AV-only data adaptation and packaged GT features. A fresh install on a single NVIDIA A10G completed all six metrics for 1,015 historical AudioX videos, followed by `aggregate` and `verify`. Compared with the prior A10G run of the same videos, KL, FAD, DeSync, IB, and CLAP matched exactly. Spatial differed by 0.0589 points because two of the 150 clips had different frame-level detector/track results; the category counts and all-150 denominator matched. Other historical models have not been rerun from this public package.
+If you use HelixBench, please cite [HelixWorld: A Real-time Interactive Audio-Visual World Model](https://arxiv.org/abs/2609.38123).

RELEASE_FILES.json

@@ -1,7 +1,7 @@
 {
   "README.md": {
-    "bytes": 4406,
-    "sha256": "a0fc046023317fda7748372d79efde6741d882e53e9ad17f6546634a69268fcb"
+    "bytes": 3311,
+    "sha256": "34a8125b7cf4c81fc81a5e07418875c306b55c0591919d5d8fd0b48511bc8b18"
   },
   "data/test.zip": {
     "bytes": 568170324,