Add RememBench evaluation inputs and documentation

100 T2V samples, 150 I2V scenes with 750 camera trajectories, 600 T2V reference annotations, preparation and validation scripts. Private release candidate; source imagery is retrieved from DL3DV by authorized users.

.gitignore

@@ -0,0 +1,4 @@
+.cache/
+__pycache__/
+*.pyc
+images/

README.md

@@ -0,0 +1,174 @@
+---
+pretty_name: RememBench
+language:
+- en
+task_categories:
+- text-to-video
+- image-to-video
+tags:
+- benchmark
+- video-generation
+- long-horizon
+- memory
+size_categories:
+- n<1K
+configs:
+- config_name: t2v
+  data_files:
+  - split: test
+    path: data/t2v/test.jsonl
+- config_name: i2v
+  data_files:
+  - split: test
+    path: data/i2v/test.jsonl
+---
+
+# RememBench
+
+RememBench evaluates consistency across long-horizon revisits in autoregressive video generation. Target objects or scenes leave the model's sliding window and later return to view. Revisits are driven by prompts in the text-to-video (T2V) split and by camera motion in the image-to-video (I2V) split.
+
+Introduced in **MosaiChunk: Compositing Spatio-Temporal Memory for Autoregressive Video Generation**.
+
+[Project page](https://mosaichunk.github.io/) · [Paper](https://mosaichunk.github.io/assets/paper.pdf) · [Video Viewer](https://mosaichunk.github.io/viewer/video.html)
+
+**Release status:** `0.1.0-rc1`, a private release candidate. The publication license is not yet assigned. This package contains the fixed evaluation inputs, camera trajectories, preparation and validation scripts, and T2V reference frame annotations and scores. DL3DV images, generated videos, model weights, and model inference implementations are not bundled.
+
+## Dataset contents
+
+| Subset | Scenes | Evaluation inputs | Input |
+|---|---:|---:|---|
+| `t2v` | 100 | 100 | Four prompt segments, their start times, and a fixed noise seed |
+| `i2v` | 150 | 750 | Scene prompt, initial-frame source, camera trajectory, and a fixed noise seed |
+
+Both subsets use the `test` split. The I2V set contains **50 indoor and 100 outdoor scenes**. All 150 scenes have 90°, 180°, and 360° rotation trajectories. The 100 outdoor scenes additionally have these three rotations combined with translation. Thus, 750 I2V rows represent 150 unique scenes; they are not 750 independent scenes.
+
+T2V scenarios are disjoint from router training. I2V conditioning frames come from DL3DV, whereas router training used Sekai frames. The evaluation trajectories were not used during training.
+
+## Load the inputs
+
+```bash
+pip install -r requirements.txt
+hf auth login
+```
+
+While the repository is private, authentication must use an account with access.
+
+```python
+from datasets import load_dataset
+
+t2v = load_dataset("evanbuzzZ/RememBench", "t2v", split="test", token=True)
+i2v = load_dataset("evanbuzzZ/RememBench", "i2v", split="test", token=True)
+
+sample = t2v[0]
+prompts = sample["prompts"]
+seed = int(sample["seed"])
+```
+
+**Seeds are decimal strings.** Some T2V seeds exceed the exact-integer range of JavaScript numbers. Convert them directly to an integer in your generation code; do not pass them through a floating-point value.
+
+Download the repository to use the trajectory files and scripts:
+
+```bash
+hf download evanbuzzZ/RememBench --repo-type dataset --local-dir RememBench
+cd RememBench
+python scripts/validate.py
+python scripts/load_example.py --subset i2v --index 0
+```
+
+For reproducible experiments, pin the Hugging Face repository commit with `revision=` or `--revision`.
+
+## T2V inputs
+
+Each row of `data/t2v/test.jsonl` contains:
+
+| Field | Meaning |
+|---|---|
+| `scene_id`, `source_index` | Stable scene identifier and original candidate index |
+| `target_description` | Target content described by the prompt |
+| `prompts` | Four prompts: appears, disappears, remains out of sight, reappears |
+| `prompt_start_seconds` | Nominal segment starts: `[0.0, 3.6, 6.2, 11.2]` |
+| `seed` | Exact noise seed, stored as a decimal string |
+| `num_frames`, `fps`, `width`, `height` | Paper setting: 379 frames, 24 fps, 1376 × 768 |
+| `num_chunks`, `denoising_steps_per_chunk` | Paper setting: 23 chunks, four denoising steps per chunk |
+
+The prompt schedule specifies intended action timing, not the exact generated action frames. The five-second third segment keeps the target out of sight longer than the largest sliding-window baseline's approximately three seconds of recent context. The paper retained 100 prompts from a 466-scene candidate pool after screening an H3-AR rollout for compliance. Evaluation uses the same retained prompts and seeds across methods and budgets.
+
+## I2V inputs and initial frames
+
+Each row of `data/i2v/test.jsonl` contains:
+
+| Field | Meaning |
+|---|---|
+| `sample_id`, `scene_id` | Unique scene–trajectory identifier and source DL3DV scene ID |
+| `scene_type` | `indoor` or `outdoor` |
+| `prompt` | Caption describing visible scene content without prescribing camera motion |
+| `image_path` | Relative path populated by the image preparation script |
+| `trajectory_path` | Relative path to a bundled NumPy archive |
+| `setting`, `rotation_degrees`, `direction`, `translation` | Camera-motion setting; direction is `left` or `right` |
+| `seed` | Paper generation seed, `"0"` |
+| `num_frames`, `fps`, `width`, `height` | Paper setting: 253 frames, 16 fps, 832 × 480 |
+| `num_chunks`, `latent_frames`, `denoising_steps_per_chunk` | 16 chunks, 64 latent frames, four denoising steps per chunk |
+| `pose_samples`, `intrinsics_samples` | Lengths of the original camera arrays |
+
+`sources/i2v_images.jsonl` contains one record per unique scene, with its upstream repository, archive path, retrieval revision, preprocessing rule, and SHA-256 of the expected RGB pixels. Initial frames are taken from the first image member in lexicographic archive-path order, center-cropped to the target aspect ratio, then resized to 832 × 480 using Lanczos.
+
+Obtain access to [DL3DV-ALL-960P](https://huggingface.co/datasets/DL3DV/DL3DV-ALL-960P) under its own terms. If you already have the scene archives or extracted scene folders:
+
+```bash
+python scripts/prepare_i2v_images.py --source-dir /path/to/DL3DV-ALL-960P
+python scripts/validate.py --require-images
+```
+
+Alternatively, retrieve the required first frames through your authorized Hugging Face account:
+
+```bash
+python scripts/prepare_i2v_images.py --download
+```
+
+The download option uses HTTP range requests to read the first image from each source archive, without downloading all of its video frames. Use `--limit 1` to try one scene first. Existing prepared images are validated and skipped. The script refuses a pixel-hash mismatch, rather than silently substituting a different input.
+
+### Camera representation
+
+Each `.npz` file has `poses` and `intrinsics`, copied without changing their numerical values from the final experiment inputs:
+
+- `poses`: float32 camera-to-world matrices of shape `(T, 4, 4)`. Camera axes are x right, y down, z forward. The source world frame is z up; the first camera is at `[0, 0, 1.5]`, facing world +y. Translation values retain the source trajectory's depth-derived scale.
+- `intrinsics`: float32 rows `[fx, fy, cx, cy]` in pixels at 832 × 480. Intrinsics are constant over each trajectory. Some source arrays have a different number of rows from `poses`; use the first intrinsics row, as in the paper's implementation.
+
+The source pose sequence contains 256 or 257 samples, not the 253 decoded video frames. In the paper's LingBot-World-Infinity setup, rotation is interpolated with SLERP and translation linearly at 64 equally spaced positions over the full pose sequence, including both endpoints. The backbone then uses relative consecutive poses with translation normalized by the largest displacement, following its camera-conditioning implementation. Preserve the complete source sequence rather than truncating it to the decoded frame count.
+
+For 90° and 180° rotation, the camera turns away and reverses. For 360°, it completes a continuous full turn. Translation moves into the scene and returns. All source trajectories end at their initial position and orientation.
+
+## Evaluation
+
+The paper uses **RAVEN-adapted MiniMax-H3 (H3-AR)** for T2V and **LingBot-World-Infinity** for I2V. Compare methods using the same prompts and noise seed; I2V methods also share the conditioning frame and input trajectory. Retrieval methods retain two recent chunks and one or two chunks of far memory. Base retains additional recent chunks to match the total active visual KV budget, with the same backbone-specific attention sink.
+
+T2V departure and revisit frames are manually annotated. The departure frame is shared across methods within a scene; the revisit is marked separately for each generated rollout. I2V uses the conditioning frame as departure and selects each rollout's revisit from its own Pi3X reconstruction. For rotation, find the frame whose viewing direction differs most from the initial direction; for translation, find the frame farthest from the initial position. From that turnaround onward, select the frame with the smallest unsigned viewing-direction angle to the initial frame. Reconstruct all decoded frames, with inputs resized to approximately 255,000 pixels and dimensions rounded to multiples of 14.
+
+| Metric | Definition | Aggregation |
+|---|---|---|
+| CLIP ↑ | Cosine similarity of L2-normalized CLIP ViT-H/14 LAION-2B embeddings of departure and revisit | Median over scenes |
+| LPIPS ↓ | AlexNet, version 0.1, on native-resolution image pairs scaled to `[-1, 1]` | Median over scenes |
+| TempSSIM ↑ | Grayscale SSIM over consecutive frames, 11 × 11 Gaussian window, sigma 1.5 | Mean within each video, then median over scenes |
+| Drift ↓ | Cosine distance between adjacent chunks, each represented by the normalized average of CLIP embeddings from four evenly spaced frames | Mean within each video, then median over scenes |
+
+H3-AR decodes an initial five-frame chunk followed by 17-frame chunks; LingBot-World-Infinity decodes an initial 13-frame chunk followed by 16-frame chunks. Scores use original rollouts before compression for the Video Viewer.
+
+`evaluation/t2v_reference_pairs.jsonl` contains 600 annotations and CLIP/LPIPS measurements: 100 scenes × three methods × two comparison budgets. Frame indices are zero-based. These annotations describe the paper's specific generated rollouts; **they are not reusable ground-truth frame indices for a new model or rerun**. Original rollout videos are not included in this input package. `source_arm` identifies the original run, and `comparison_budget_chunks` identifies the matched-budget comparison; Base has no retrieved far memory.
+
+`evaluation/t2v_reference_summary.json` reproduces the paper's median T2V CLIP and LPIPS values:
+
+| Method | CLIP, 1 chunk | LPIPS, 1 chunk | CLIP, 2 chunks | LPIPS, 2 chunks |
+|---|---:|---:|---:|---:|
+| Base | 0.755 | 0.659 | 0.751 | 0.652 |
+| MoC | 0.832 | 0.594 | 0.841 | 0.569 |
+| MosaiChunk | 0.899 | 0.567 | 0.936 | 0.500 |
+
+## Provenance, limitations, and licensing
+
+`sources/input_provenance.json` records hashes of the source manifests, prompts, trajectories, seeds, and annotations used for this export. `checksums.sha256` covers the distributed release files.
+
+Prompt compliance screening used one H3-AR rollout per candidate; success rates can differ with another model. T2V revisit annotation is manual, and I2V revisit selection depends on reconstructed poses. Report the evaluated cohort and any generation failures rather than silently changing the test set. The dataset measures revisits in the supplied scenarios and trajectories; it is not an exhaustive measure of video quality or world understanding.
+
+Publication licensing for the RememBench-authored material is pending. DL3DV content remains subject to the [DL3DV license and Terms of Use](https://github.com/DL3DV-10K/Dataset/blob/main/License.md); this repository does not grant additional rights to it. The initial-frame pixels are obtained from the upstream source by the user and are not redistributed here. The four-stage T2V design extends the appear–disappear–reappear design of Ring Forcing; construction details and citations are in the linked paper.
+
+Please cite the MosaiChunk paper when using RememBench and acknowledge DL3DV when using the I2V split. Canonical paper citation metadata will be added with the public release.

evaluation/t2v_reference_summary.json

@@ -0,0 +1,44 @@
+[
+  {
+    "method": "Base",
+    "comparison_budget_chunks": 1,
+    "n": 100,
+    "median_clip": 0.755422979593277,
+    "median_lpips_alex": 0.6593886017799377
+  },
+  {
+    "method": "MoC",
+    "comparison_budget_chunks": 1,
+    "n": 100,
+    "median_clip": 0.831767350435257,
+    "median_lpips_alex": 0.5944797098636627
+  },
+  {
+    "method": "MosaiChunk",
+    "comparison_budget_chunks": 1,
+    "n": 100,
+    "median_clip": 0.8989355564117432,
+    "median_lpips_alex": 0.5666199922561646
+  },
+  {
+    "method": "Base",
+    "comparison_budget_chunks": 2,
+    "n": 100,
+    "median_clip": 0.7511980831623077,
+    "median_lpips_alex": 0.6515471041202545
+  },
+  {
+    "method": "MoC",
+    "comparison_budget_chunks": 2,
+    "n": 100,
+    "median_clip": 0.8408968150615692,
+    "median_lpips_alex": 0.5687620043754578
+  },
+  {
+    "method": "MosaiChunk",
+    "comparison_budget_chunks": 2,
+    "n": 100,
+    "median_clip": 0.9362636506557465,
+    "median_lpips_alex": 0.4996824115514755
+  }
+]

release_info.json

@@ -0,0 +1,18 @@
+{
+  "version": "0.1.0-rc1",
+  "status": "private release candidate",
+  "t2v_samples": 100,
+  "i2v_scenes": 150,
+  "i2v_samples": 750,
+  "t2v_reference_pairs": 600,
+  "source_images_included": false,
+  "publication_license": "pending",
+  "trajectory_settings": {
+    "rotation_90": 150,
+    "rotation_180": 150,
+    "rotation_360": 150,
+    "translation_90": 100,
+    "translation_180": 100,
+    "translation_360": 100
+  }
+}

requirements.txt

@@ -0,0 +1,5 @@
+datasets>=4.0
+huggingface_hub>=1.0
+numpy>=1.24
+Pillow>=10
+fsspec[http]>=2023.1

scripts/load_example.py

@@ -0,0 +1,28 @@
+#!/usr/bin/env python3
+"""Inspect a sample from a downloaded copy of RememBench."""
+import argparse
+import json
+from pathlib import Path
+
+import numpy as np
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument('--root', type=Path, default=Path(__file__).resolve().parents[1])
+    ap.add_argument('--subset', choices=['t2v', 'i2v'], default='t2v')
+    ap.add_argument('--index', type=int, default=0)
+    args = ap.parse_args()
+    rows = [json.loads(line) for line in (args.root / 'data' / args.subset / 'test.jsonl').read_text().splitlines()]
+    row = rows[args.index]
+    print(json.dumps(row, indent=2))
+    print('Exact integer seed:', int(row['seed']))
+    if args.subset == 'i2v':
+        with np.load(args.root / row['trajectory_path'], allow_pickle=False) as trajectory:
+            print('Camera-to-world poses:', trajectory['poses'].shape)
+            print('Intrinsics [fx, fy, cx, cy]:', trajectory['intrinsics'][0])
+        print('Conditioning image prepared:', (args.root / row['image_path']).is_file())
+
+
+if __name__ == '__main__':
+    main()

scripts/prepare_i2v_images.py

@@ -0,0 +1,97 @@
+#!/usr/bin/env python3
+"""Prepare the exact I2V inputs from user-authorized DL3DV source data."""
+import argparse
+import hashlib
+import io
+import json
+from pathlib import Path, PurePosixPath
+import zipfile
+
+from PIL import Image
+
+EXTENSIONS = {'.png', '.jpg', '.jpeg'}
+
+
+def prepare_image(data, row):
+    image = Image.open(io.BytesIO(data)).convert('RGB')
+    w, h = image.size
+    width, height = row['width'], row['height']
+    aspect = width / height
+    if w / h > aspect:
+        crop = int(round(h * aspect))
+        image = image.crop(((w - crop) // 2, 0, (w - crop) // 2 + crop, h))
+    elif w / h < aspect:
+        crop = int(round(w / aspect))
+        image = image.crop((0, (h - crop) // 2, w, (h - crop) // 2 + crop))
+    image = image.resize((width, height), Image.Resampling.LANCZOS)
+    actual = hashlib.sha256(image.tobytes()).hexdigest()
+    if actual != row['rgb_sha256']:
+        raise ValueError(f"Input image mismatch for {row['scene_id']}: expected {row['rgb_sha256']}, got {actual}. Check the upstream source and preprocessing version.")
+    return image
+
+
+def archive_image(archive):
+    with zipfile.ZipFile(archive) as z:
+        names = sorted(n for n in z.namelist() if PurePosixPath(n).suffix.lower() in EXTENSIONS)
+        if not names:
+            raise ValueError(f'No image members in {archive}')
+        return z.read(names[0])
+
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument('--root', type=Path, default=Path(__file__).resolve().parents[1])
+    source = ap.add_mutually_exclusive_group(required=True)
+    source.add_argument('--source-dir', type=Path, help='Local DL3DV archives or extracted scene directories')
+    source.add_argument('--download', action='store_true', help='Read first frames from DL3DV with HTTP range requests using your own HF access')
+    ap.add_argument('--limit', type=int, help='Prepare only the first N scenes')
+    args = ap.parse_args()
+    if args.limit is not None and args.limit < 1:
+        ap.error('--limit must be positive')
+    rows = [json.loads(s) for s in (args.root / 'sources/i2v_images.jsonl').read_text().splitlines()]
+    if args.limit is not None:
+        rows = rows[:args.limit]
+    archives = {}
+    if args.source_dir:
+        if not args.source_dir.is_dir():
+            ap.error('--source-dir must be an existing directory')
+        for path in args.source_dir.rglob('*.zip'):
+            if path.stem in archives:
+                raise ValueError(f'Ambiguous source archives for {path.stem}')
+            archives[path.stem] = path
+    for i, row in enumerate(rows, 1):
+        target = args.root / row['image_path']
+        if target.is_file():
+            im = Image.open(target).convert('RGB')
+            if im.size != (row['width'], row['height']) or hashlib.sha256(im.tobytes()).hexdigest() != row['rgb_sha256']:
+                raise ValueError(f'Existing image does not match the benchmark: {target}')
+            print(f'{i}/{len(rows)} verified {row["scene_id"]}', flush=True)
+            continue
+        if args.download:
+            import fsspec
+            from huggingface_hub import get_token, hf_hub_url
+            token = get_token()
+            if not token:
+                raise RuntimeError('Run hf auth login with an account that has accepted the DL3DV terms.')
+            url = hf_hub_url(repo_id=row['source_repo'], filename=row['source_archive'],
+                             revision=row['source_revision'], repo_type='dataset')
+            with fsspec.open(url, mode='rb', block_size=65536,
+                             headers={'Authorization': 'Bearer ' + token}) as remote:
+                data = archive_image(remote)
+        elif row['scene_id'] in archives:
+            data = archive_image(archives[row['scene_id']])
+        else:
+            scene = args.source_dir / row['scene_id']
+            files = sorted((p for p in scene.rglob('*') if p.is_file() and p.suffix.lower() in EXTENSIONS),
+                           key=lambda p: p.relative_to(scene).as_posix())
+            if not files:
+                raise FileNotFoundError(f"No archive or extracted images found for {row['scene_id']}")
+            data = files[0].read_bytes()
+        image = prepare_image(data, row)
+        target.parent.mkdir(parents=True, exist_ok=True)
+        image.save(target)
+        print(f'{i}/{len(rows)} prepared {row["scene_id"]}', flush=True)
+
+
+if __name__ == '__main__':
+    main()

scripts/validate.py

@@ -0,0 +1,93 @@
+#!/usr/bin/env python3
+"""Check release integrity, cohorts, camera arrays, and reference statistics."""
+import argparse
+from collections import Counter
+import hashlib
+import json
+from pathlib import Path
+
+import numpy as np
+
+
+def check(condition, message):
+    if not condition:
+        raise ValueError(message)
+
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument('--root', type=Path, default=Path(__file__).resolve().parents[1])
+    ap.add_argument('--require-images', action='store_true')
+    args = ap.parse_args()
+    root = args.root
+    for line in (root / 'checksums.sha256').read_text().splitlines():
+        expected, relative = line.split('  ', 1)
+        path = root / relative
+        check(path.is_file(), f'Missing release file: {relative}')
+        check(hashlib.sha256(path.read_bytes()).hexdigest() == expected, f'Checksum mismatch: {relative}')
+
+    def rows(name):
+        return [json.loads(s) for s in (root / name).read_text().splitlines()]
+
+    t2v, i2v = rows('data/t2v/test.jsonl'), rows('data/i2v/test.jsonl')
+    sources = rows('sources/i2v_images.jsonl')
+    check(len(t2v) == len({r['scene_id'] for r in t2v}) == 100, 'T2V scene count or uniqueness mismatch')
+    check(len(i2v) == len({r['sample_id'] for r in i2v}) == 750, 'I2V sample count or uniqueness mismatch')
+    check(len(sources) == len({r['scene_id'] for r in i2v}) == 150, 'I2V scene count mismatch')
+    check(Counter(r['scene_type'] for r in sources) == {'indoor': 50, 'outdoor': 100}, 'Incorrect scene mix')
+    for row in t2v + i2v:
+        check(isinstance(row['seed'], str) and row['seed'].isdigit(), 'Seed must be a decimal string')
+        check(0 <= int(row['seed']) < 2**63, 'Seed out of range')
+    for row in t2v:
+        check(len(row['prompts']) == 4 and all(row['prompts']), 'Incomplete T2V prompt schedule')
+        check(row['prompt_start_seconds'] == [0.0, 3.6, 6.2, 11.2], 'T2V prompt timing mismatch')
+    cohorts = {}
+    for row in i2v:
+        cohorts.setdefault(row['setting'], set()).add(row['scene_id'])
+        with np.load(root / row['trajectory_path'], allow_pickle=False) as data:
+            p, k = data['poses'], data['intrinsics']
+            check(p.shape == (row['pose_samples'], 4, 4), 'Pose shape mismatch')
+            check(k.shape == (row['intrinsics_samples'], 4), 'Intrinsics shape mismatch')
+            check(np.isfinite(p).all() and np.isfinite(k).all(), 'Non-finite camera values')
+            check(np.allclose(k, k[0]), 'Expected constant camera intrinsics')
+            check(np.allclose(p[0], p[-1], atol=1e-5), 'Trajectory does not return to its origin')
+            check(np.allclose(p[:, 3], [0, 0, 0, 1]), 'Invalid homogeneous pose')
+            r = p[:, :3, :3]
+            check(np.allclose(r.transpose(0, 2, 1) @ r, np.eye(3), atol=1e-5), 'Invalid rotation matrix')
+            check(np.allclose(np.linalg.det(r), 1, atol=1e-5), 'Invalid rotation determinant')
+            moving = np.max(np.linalg.norm(p[:, :3, 3] - p[0, :3, 3], axis=1)) > 1e-5
+            check(bool(moving) == row['translation'], 'Translation flag disagrees with camera path')
+    all_scenes = {r['scene_id'] for r in sources}
+    outdoors = {r['scene_id'] for r in sources if r['scene_type'] == 'outdoor'}
+    for angle in [90, 180, 360]:
+        check(cohorts[f'rotation_{angle}'] == all_scenes, 'Rotation cohort mismatch')
+        check(cohorts[f'translation_{angle}'] == outdoors, 'Translation cohort mismatch')
+    pairs = rows('evaluation/t2v_reference_pairs.jsonl')
+    check(len(pairs) == 600, 'Missing reference annotations')
+    check(len({(p['scene_id'], p['method'], p['comparison_budget_chunks']) for p in pairs}) == 600,
+          'Duplicate reference annotations')
+    for p in pairs:
+        check(0 <= p['departure_frame'] < p['revisit_frame'] < 379, 'Reference frame out of range')
+    summary = json.loads((root / 'evaluation/t2v_reference_summary.json').read_text())
+    for s in summary:
+        subset = [p for p in pairs if p['method'] == s['method'] and p['comparison_budget_chunks'] == s['comparison_budget_chunks']]
+        check(len(subset) == s['n'] == 100, 'Reference cohort mismatch')
+        for key in ['clip', 'lpips_alex']:
+            check(np.isclose(np.median([p[key] for p in subset]), s[f'median_{key}']), 'Reference summary mismatch')
+    prepared = 0
+    from PIL import Image
+    for row in sources:
+        path = root / row['image_path']
+        if args.require_images:
+            check(path.is_file(), f'Missing conditioning image: {path}')
+        if path.is_file():
+            image = Image.open(path).convert('RGB')
+            check(image.size == (832, 480), 'Conditioning image dimensions differ')
+            check(hashlib.sha256(image.tobytes()).hexdigest() == row['rgb_sha256'], 'Conditioning image differs')
+            prepared += 1
+    print(json.dumps(dict(status='passed', t2v_samples=100, i2v_scenes=150, i2v_samples=750,
+                          trajectory_files=750, t2v_reference_pairs=600, prepared_images=prepared), indent=2))
+
+
+if __name__ == '__main__':
+    main()

trajectories/rotation_180/006771db3c057280f9277e735be6daa24339657ce999216c38da68002a443fed.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/00b1ad87c296635c73bbc9728a63d3df4a9ab07aee0d34cd45784e39f2c699ba.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/00e3fa12dff0aa4791ee9077de752c29a77e7b64c809c326e93f13df23c84875.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/02edaedb5016e1c5c9d9746275e8acdcb78b38fda68b051a6b29271b0ae131de.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/03c53f58fc38f5bce2cb4301261711bc43469905c2eeca17e3fd9a6556bd6a88.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/0742f6f08fec5b4ad92c1720fa53408780a8adb4dad0e92c8817a4f46c45074b.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/079fa662f29b06eda156a5f76255541aa2581e0bccd257c6a48103f271cea191.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/07dea22dcbe1fbd7a85370a48091cd46cdf395a81464de82c7f7d71461973518.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/089c0262673d945f7e333635321ce6d6065238ce96ae50c7f36d078cf15b601f.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/0a45aa466f114e6daa13a82775e7bd5fc295e37a1e3d61deb3741a7e7a1b1f8a.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/0de7a53aad7ddffd26f7db5d3f7ce61d27b3344de2b037289443259da7dac805.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/0e24a8c5a728a4d0f1d3d131ca339a93da01b119e6c01bb7b290fad3a201816f.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/135fc2a1c31c80698903deec6728067ed03de4f600df0a0f530d8f72cc5bb821.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/17281837e808c89f1690dd0a141a9050cad1d7725c7abd38947c51712b0498a9.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/17abed01812ec9ba0d42c901315e9fc0b4b338eb18a0ce6653fbec2bacd87f76.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/183128db642a3f0841ef790cc5fed75b014ffeb2760a8925793aea5ffb89693d.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/1a0dc4cda12b60b6eb02464b8164944a6fd28bf92341e46b5ca1271526e4c0ad.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/1b4b05e87c7c6402ef47ef8c723e11605048bba0ffa81d3e834e79ef18415d3e.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/1b4f0f20dc2187e5004d04f3cb0082ee1b7a57b2c54ce8d5b2fe3888816d9b62.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/1b5417305c2de43c8c091c27363707106dcf681c1c4ab0cf704c6d13070c0954.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/1ddddfa75644cb9651fa1885482c8836890cbd4792d4e5401c9c9dfaac6fc1dc.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/1de3570f4006258a1ed396236afe779123d51ec412646bad95c53f63f64ded3c.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/1e7324084a5fa344ec9df616af0bbb854d554f960dd43dd4089445e7cc729d52.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/1eda90dc0c9fcd24d2a333d0eb3cf4caa411acc1820e931583a587d94c70aa33.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/212b69281c7f18265442f4aa66ce7989fa9e99af6c465e1f2af09645caacf464.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/2370a8bb7de3119db312334d841dc589a1a4f308532366ee836fd79488030d54.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/239db1d99c9ec1ff1a593288d30b1c516adcf3e1326b6151685acaa3176bba5e.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/23fbc88477094ec18cd666b1c6781b69d6d265a5fd3be71f8469569043188636.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/24ee1626d30ead0e47d6b75a44fb70dc1847aa73e48ca75592d75672c24a0da9.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/24febed2ccef973f672c3ec3d9b26073b6f0cb72fc674288b3591e6eb93cf9d4.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/25ef179c7c3c5f506406c86c22ae9f7c9184b32cdd44a19831cd109a65c207fc.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/26c5bcf6b7d63cc5618d38c65d699a89795038f0a7fc054b8ce17ab86e01e328.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cce029ed3d8bf9b20cc696a78dadb682a16f7e7e35d0698051eeb0ff8ad21aab
+size 2685

trajectories/rotation_180/279e2c7fa97886715b7698fd7bd59af5b0aaf5e89b36cbe43214cc4cd341d553.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/2a1f802bfa0cff9c6969fb3adc7656320cbd2e6105ee91c263a6fefd04d8bf9e.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/30a3635f537837fbc8d68870144ef1a800a9939d4ed524f5e943ba2f80fe7019.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666

trajectories/rotation_180/3648c25b94391237b8c2d0d851dba3ca6e33d4bc9c4c2b4b8fcf8141ba2f2542.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28e5e780af8e7c3e43f06f6351373ac426f04efceba604eb4f8bf8aa089404eb
+size 2666