Initial Commit

.gitattributes

@@ -0,0 +1,60 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,116 @@
+---
+viewer: false
+---
+
+# ConfigWM Dataset
+
+Data for [**World Models with Configurable Perception**](https://arxiv.org/abs/2610.27914), covering **Push-T**, **Two-Room**, and **OGBench-Cube** with distraction variants `d0` through `d8`.
+
+The dataset supports query-conditioned visual representation learning, action-conditioned world modeling, and online control evaluation.
+
+## Environments
+
+`d0` has no distractors; `d1` through `d8` add one through eight. Click the figure to view the original image.
+
+[![Two-Room, Push-T, and OGBench-Cube with zero through eight distractors](assets/environments/noisy_world.png)](assets/environments/noisy_world.png)
+
+## Data
+
+Each environment uses the same three collections:
+
+| Collection | Contents | Use |
+| --- | --- | --- |
+| `trajectories/` | RGB observations, actions, episode metadata, and queries | Training ConfigWM |
+| `perception/` | Source images, text queries, and target images | Training Configurable Encoder |
+| `control/` | Reset trajectories and references to shared trajectory data | Online evaluation |
+
+```text
+dataset.json
+assets/environments/       # README previews
+pusht/                     # also tworoom/ and ogbcube/
+  trajectories/d0...d8/
+    manifest.json
+    queries.jsonl
+    shard_*/frames.h5
+    shard_*/episodes.jsonl
+  perception/
+    manifest.json
+    shard_*/pairs.h5
+  control/
+    d0...d8/manifest.json
+    base.h5                # Push-T and Two-Room
+```
+
+### Format
+
+| Data | Format |
+| --- | --- |
+| RGB observations | HDF5, `uint8`, `[frame, height, width, channel]` |
+| Actions | 2 dimensions for Push-T/Two-Room; 5 for OGBench-Cube |
+| Perception pairs | HDF5 `source_pixels`, `target_pixels`, and `query` |
+| Descriptions | JSONL episode metadata and training queries |
+
+- `dataset.json`: entry manifests for each environment and task.
+- `field_map`: HDF5 file role and array key for each field.
+- `ep_offset`: starting frame index of each episode.
+- `ep_len`: number of frames in each episode.
+
+## Download
+
+The full dataset is approximately **1.64 TB**. To download one environment and variant, run this from the ConfigWM code repository:
+
+```bash
+python tools/download_data.py \
+  --repo YijunShen/ConfigWM \
+  --local-dir ./configwm-data \
+  --env pusht --task stage_b --variant d3
+```
+
+- `--env`: `pusht`, `tworoom`, or `ogbcube`.
+- `--task stage_b --variant d3`: data for training ConfigWM on one variant.
+- `--task stage_a`: data for training Configurable Encoder in one environment.
+- `--task control --variant d3`: evaluation resources and their dependencies.
+- `--list-only`: fetch small manifests and list the selected files.
+
+Reuse the same local directory across downloads to share common files.
+
+To download everything:
+
+```python
+from huggingface_hub import snapshot_download
+
+snapshot_download(
+    repo_id="YijunShen/ConfigWM",
+    repo_type="dataset",
+    local_dir="configwm-data",
+)
+```
+
+## Real-world perception
+
+The [real-world collection](https://huggingface.co/datasets/YijunShen/ConfigWM/tree/main/realworld) contains image pairs with task descriptions.
+
+```bash
+hf download YijunShen/ConfigWM --repo-type dataset \
+  --include "realworld/**" --local-dir ./configwm-data
+```
+
+Use `./configwm-data/realworld` as `--data-root` with `--env realworld`. See the [format and usage](https://huggingface.co/datasets/YijunShen/ConfigWM/blob/main/realworld/README.md).
+
+## Citation
+
+```bibtex
+@misc{shen2026configwm,
+      title={World Models with Configurable Perception},
+      author={Yijun Shen and Delong Chen and Xianming Hu and Kai Zhang and Pascale Fung},
+      year={2026},
+      eprint={2610.27914},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2610.27914},
+}
+```
+
+## Acknowledgments
+
+We would like to thank Willy Chung, Quentin Garrido, Basile Terver for providing valuable feedback to this work.

ckpts/ogbcube/configs/data_config.yaml

@@ -0,0 +1,4 @@
+preprocessor:
+  id: dinov2
+  model_name: facebook/dinov2-small
+  resolution: 224

ckpts/ogbcube/configs/model_config.yaml

@@ -0,0 +1,36 @@
+vision_embedding:
+  type: dinov2
+  model_name: facebook/dinov2-small
+
+predictor:
+  backbone:
+    type: llama3
+    model_name: alpindale/Llama-3.2-1B
+    pretrained_weights: true
+    dtype: float32
+    start_layer: 0
+    num_layers: 3
+  max_input_tokens: 257
+  max_query_tokens: 32
+  pooling: patch_tokens
+
+text_encoder:
+  disabled: true
+
+projector:
+  type: linear
+  dim: 384
+
+freeze_vision_encoder: false
+separate_target_vision_encoder: true
+freeze_target_vision_encoder: true
+freeze_predictor_token_embeddings: true
+use_target_vision_projector: false
+add_query_to_target: false
+use_sequence_packing: false
+target_vision_pooling: patch_tokens
+
+# Patch-position InfoNCE, computed by the trainer.
+same_source_contrastive_weight: 1.0
+same_source_contrastive_temperature: 0.07
+same_source_contrastive_detach_target: true

ckpts/ogbcube/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90e2f078ea96abbcad910723b6b27d150a9dcc88049a17ff028515d1f84ebb52
+size 1963357908

ckpts/pusht/configs/data_config.yaml

@@ -0,0 +1,4 @@
+preprocessor:
+  id: dinov2
+  model_name: facebook/dinov2-small
+  resolution: 224

ckpts/pusht/configs/model_config.yaml

@@ -0,0 +1,36 @@
+vision_embedding:
+  type: dinov2
+  model_name: facebook/dinov2-small
+
+predictor:
+  backbone:
+    type: llama3
+    model_name: alpindale/Llama-3.2-1B
+    pretrained_weights: true
+    dtype: float32
+    start_layer: 0
+    num_layers: 3
+  max_input_tokens: 257
+  max_query_tokens: 32
+  pooling: patch_tokens
+
+text_encoder:
+  disabled: true
+
+projector:
+  type: linear
+  dim: 384
+
+freeze_vision_encoder: false
+separate_target_vision_encoder: true
+freeze_target_vision_encoder: true
+freeze_predictor_token_embeddings: true
+use_target_vision_projector: false
+add_query_to_target: false
+use_sequence_packing: false
+target_vision_pooling: patch_tokens
+
+# Patch-position InfoNCE, computed by the trainer.
+same_source_contrastive_weight: 1.0
+same_source_contrastive_temperature: 0.07
+same_source_contrastive_detach_target: true

ckpts/pusht/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd647cb3cbf63498f38debccaa1b1c38ac2e86108f7f8e85fe6c6e6f9aebf374
+size 1963357908

ckpts/tworoom/configs/data_config.yaml

@@ -0,0 +1,4 @@
+preprocessor:
+  id: dinov2
+  model_name: facebook/dinov2-small
+  resolution: 224

ckpts/tworoom/configs/model_config.yaml

@@ -0,0 +1,36 @@
+vision_embedding:
+  type: dinov2
+  model_name: facebook/dinov2-small
+
+predictor:
+  backbone:
+    type: llama3
+    model_name: alpindale/Llama-3.2-1B
+    pretrained_weights: true
+    dtype: float32
+    start_layer: 0
+    num_layers: 3
+  max_input_tokens: 257
+  max_query_tokens: 32
+  pooling: patch_tokens
+
+text_encoder:
+  disabled: true
+
+projector:
+  type: linear
+  dim: 384
+
+freeze_vision_encoder: false
+separate_target_vision_encoder: true
+freeze_target_vision_encoder: true
+freeze_predictor_token_embeddings: true
+use_target_vision_projector: false
+add_query_to_target: false
+use_sequence_packing: false
+target_vision_pooling: patch_tokens
+
+# Patch-position InfoNCE, computed by the trainer.
+same_source_contrastive_weight: 1.0
+same_source_contrastive_temperature: 0.07
+same_source_contrastive_detach_target: true

ckpts/tworoom/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38974a2e25fb06f98e7ce8924c590e3821e2c23253a997dffb077bdb20dde0eb
+size 1963357908

ckpts/unified/configs/data_config.yaml

@@ -0,0 +1,4 @@
+preprocessor:
+  id: dinov2
+  model_name: facebook/dinov2-small
+  resolution: 224

ckpts/unified/configs/model_config.yaml

@@ -0,0 +1,36 @@
+vision_embedding:
+  type: dinov2
+  model_name: facebook/dinov2-small
+
+predictor:
+  backbone:
+    type: llama3
+    model_name: alpindale/Llama-3.2-1B
+    pretrained_weights: true
+    dtype: float32
+    start_layer: 0
+    num_layers: 3
+  max_input_tokens: 257
+  max_query_tokens: 32
+  pooling: patch_tokens
+
+text_encoder:
+  disabled: true
+
+projector:
+  type: linear
+  dim: 384
+
+freeze_vision_encoder: false
+separate_target_vision_encoder: true
+freeze_target_vision_encoder: true
+freeze_predictor_token_embeddings: true
+use_target_vision_projector: false
+add_query_to_target: false
+use_sequence_packing: false
+target_vision_pooling: patch_tokens
+
+# Patch-position InfoNCE, computed by the trainer.
+same_source_contrastive_weight: 1.0
+same_source_contrastive_temperature: 0.07
+same_source_contrastive_detach_target: true

ckpts/unified/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1fb6fb0a0a263bd361c1f777b88a635eea8cd5c095f23c3f007bee82456757e
+size 1963357908

dataset.json

@@ -0,0 +1,110 @@
+{
+  "pusht": {
+    "stage_a": [
+      "pusht/trajectories/d0/manifest.json",
+      "pusht/trajectories/d1/manifest.json",
+      "pusht/trajectories/d2/manifest.json",
+      "pusht/trajectories/d3/manifest.json",
+      "pusht/trajectories/d4/manifest.json",
+      "pusht/trajectories/d5/manifest.json",
+      "pusht/trajectories/d6/manifest.json",
+      "pusht/trajectories/d7/manifest.json",
+      "pusht/trajectories/d8/manifest.json",
+      "pusht/perception/manifest.json"
+    ],
+    "stage_b": {
+      "d0": "pusht/trajectories/d0/manifest.json",
+      "d1": "pusht/trajectories/d1/manifest.json",
+      "d2": "pusht/trajectories/d2/manifest.json",
+      "d3": "pusht/trajectories/d3/manifest.json",
+      "d4": "pusht/trajectories/d4/manifest.json",
+      "d5": "pusht/trajectories/d5/manifest.json",
+      "d6": "pusht/trajectories/d6/manifest.json",
+      "d7": "pusht/trajectories/d7/manifest.json",
+      "d8": "pusht/trajectories/d8/manifest.json"
+    },
+    "control": {
+      "d0": "pusht/control/d0/manifest.json",
+      "d1": "pusht/control/d1/manifest.json",
+      "d2": "pusht/control/d2/manifest.json",
+      "d3": "pusht/control/d3/manifest.json",
+      "d4": "pusht/control/d4/manifest.json",
+      "d5": "pusht/control/d5/manifest.json",
+      "d6": "pusht/control/d6/manifest.json",
+      "d7": "pusht/control/d7/manifest.json",
+      "d8": "pusht/control/d8/manifest.json"
+    }
+  },
+  "tworoom": {
+    "stage_a": [
+      "tworoom/trajectories/d0/manifest.json",
+      "tworoom/trajectories/d1/manifest.json",
+      "tworoom/trajectories/d2/manifest.json",
+      "tworoom/trajectories/d3/manifest.json",
+      "tworoom/trajectories/d4/manifest.json",
+      "tworoom/trajectories/d5/manifest.json",
+      "tworoom/trajectories/d6/manifest.json",
+      "tworoom/trajectories/d7/manifest.json",
+      "tworoom/trajectories/d8/manifest.json",
+      "tworoom/perception/manifest.json"
+    ],
+    "stage_b": {
+      "d0": "tworoom/trajectories/d0/manifest.json",
+      "d1": "tworoom/trajectories/d1/manifest.json",
+      "d2": "tworoom/trajectories/d2/manifest.json",
+      "d3": "tworoom/trajectories/d3/manifest.json",
+      "d4": "tworoom/trajectories/d4/manifest.json",
+      "d5": "tworoom/trajectories/d5/manifest.json",
+      "d6": "tworoom/trajectories/d6/manifest.json",
+      "d7": "tworoom/trajectories/d7/manifest.json",
+      "d8": "tworoom/trajectories/d8/manifest.json"
+    },
+    "control": {
+      "d0": "tworoom/control/d0/manifest.json",
+      "d1": "tworoom/control/d1/manifest.json",
+      "d2": "tworoom/control/d2/manifest.json",
+      "d3": "tworoom/control/d3/manifest.json",
+      "d4": "tworoom/control/d4/manifest.json",
+      "d5": "tworoom/control/d5/manifest.json",
+      "d6": "tworoom/control/d6/manifest.json",
+      "d7": "tworoom/control/d7/manifest.json",
+      "d8": "tworoom/control/d8/manifest.json"
+    }
+  },
+  "ogbcube": {
+    "stage_a": [
+      "ogbcube/trajectories/d0/manifest.json",
+      "ogbcube/trajectories/d1/manifest.json",
+      "ogbcube/trajectories/d2/manifest.json",
+      "ogbcube/trajectories/d3/manifest.json",
+      "ogbcube/trajectories/d4/manifest.json",
+      "ogbcube/trajectories/d5/manifest.json",
+      "ogbcube/trajectories/d6/manifest.json",
+      "ogbcube/trajectories/d7/manifest.json",
+      "ogbcube/trajectories/d8/manifest.json",
+      "ogbcube/perception/manifest.json"
+    ],
+    "stage_b": {
+      "d0": "ogbcube/trajectories/d0/manifest.json",
+      "d1": "ogbcube/trajectories/d1/manifest.json",
+      "d2": "ogbcube/trajectories/d2/manifest.json",
+      "d3": "ogbcube/trajectories/d3/manifest.json",
+      "d4": "ogbcube/trajectories/d4/manifest.json",
+      "d5": "ogbcube/trajectories/d5/manifest.json",
+      "d6": "ogbcube/trajectories/d6/manifest.json",
+      "d7": "ogbcube/trajectories/d7/manifest.json",
+      "d8": "ogbcube/trajectories/d8/manifest.json"
+    },
+    "control": {
+      "d0": "ogbcube/control/d0/manifest.json",
+      "d1": "ogbcube/control/d1/manifest.json",
+      "d2": "ogbcube/control/d2/manifest.json",
+      "d3": "ogbcube/control/d3/manifest.json",
+      "d4": "ogbcube/control/d4/manifest.json",
+      "d5": "ogbcube/control/d5/manifest.json",
+      "d6": "ogbcube/control/d6/manifest.json",
+      "d7": "ogbcube/control/d7/manifest.json",
+      "d8": "ogbcube/control/d8/manifest.json"
+    }
+  }
+}

ogbcube/control/d0/manifest.json

@@ -0,0 +1,6 @@
+{
+  "shards": [],
+  "dependencies": [
+    "ogbcube/trajectories/d0/manifest.json"
+  ]
+}

ogbcube/control/d1/manifest.json

@@ -0,0 +1,7 @@
+{
+  "shards": [],
+  "dependencies": [
+    "ogbcube/trajectories/d0/manifest.json",
+    "ogbcube/trajectories/d1/manifest.json"
+  ]
+}

ogbcube/control/d2/manifest.json

@@ -0,0 +1,7 @@
+{
+  "shards": [],
+  "dependencies": [
+    "ogbcube/trajectories/d0/manifest.json",
+    "ogbcube/trajectories/d2/manifest.json"
+  ]
+}

ogbcube/control/d3/manifest.json

@@ -0,0 +1,7 @@
+{
+  "shards": [],
+  "dependencies": [
+    "ogbcube/trajectories/d0/manifest.json",
+    "ogbcube/trajectories/d3/manifest.json"
+  ]
+}

ogbcube/control/d4/manifest.json

@@ -0,0 +1,7 @@
+{
+  "shards": [],
+  "dependencies": [
+    "ogbcube/trajectories/d0/manifest.json",
+    "ogbcube/trajectories/d4/manifest.json"
+  ]
+}

ogbcube/control/d5/manifest.json

@@ -0,0 +1,7 @@
+{
+  "shards": [],
+  "dependencies": [
+    "ogbcube/trajectories/d0/manifest.json",
+    "ogbcube/trajectories/d5/manifest.json"
+  ]
+}