CJEPA
This repository contains the weights and code for Causal-JEPA (C-JEPA), a simple and flexible object-centric world model architecture presented in the paper Causal-JEPA: Learning World Models through Object-Level...
Model source
- Kind
- Other
- File size
- 13.6–133 MB
- Version
- vRevision d56bfc1
- Updated
- Nov 2, 2026
- Downloads
- 0
- Source
- Hugging Face
Source description
C-JEPA: Causal-JEPA
This repository contains the weights and code for Causal-JEPA (C-JEPA), a simple and flexible object-centric world model architecture presented in the paper Causal-JEPA: Learning World Models through Object-Level Latent Interventions.
- Paper: Causal-JEPA: Learning World Models through Object-Level Latent Interventions
- Project Page: https://hazel-heejeong-nam.github.io/cjepa/
- Code: https://github.com/galilai-group/cjepa
Summary
World models require robust relational understanding to support prediction, reasoning, and control. While object-centric representations provide a useful abstraction, they are not sufficient to capture interaction-dependent dynamics. C-JEPA is a simple and flexible object-centric world model that extends masked joint embedding prediction from image patches to object-centric representations. By applying object-level masking that requires an object's state to be inferred from other objects, C-JEPA induces latent interventions with counterfactual-like effects and prevents shortcut solutions, making interaction reasoning essential.
Empirically, C-JEPA demonstrates:
- Improved Visual Reasoning: Consistent gains in visual question answering, with an absolute improvement of about 20% in counterfactual reasoning compared to the same architecture without object-level masking on benchmarks like CLEVRER.
- Efficient Planning: Substantially more efficient planning in agent control tasks (e.g., Push-T), using only 1% of the total latent input features required by patch-based world models while achieving comparable performance.
- Causal Inductive Bias: A formal analysis demonstrates that object-level masking induces a causal inductive bias via latent interventions.
Architecture
architecture
Setup and Usage
C-JEPA relies on object-centric encoders (like VideoSAUR or SAVi) to extract representations. For detailed environment setup, dataset preparation, and training/evaluation scripts, please refer to the official GitHub repository. The repository also provides model checkpoints and pre-extracted slot representations for various configurations.
Citation
If you find this work useful, please consider citing:
@article{nam2026causal,
title={Causal-JEPA: Learning World Models through Object-Level Latent Interventions},
author={Nam, Heejeong and Le Lidec, Quentin and Maes, Lucas and LeCun, Yann and Balestriero, Randall},
journal={arXiv preprint arXiv:2602.11389},
year={2026}
}
Sources
1 source
Sources
1 sourceVerified Nov 2
Download files
Repository files
File inventory only; completeness is unverified.
- cjepa-ckpts/clevrer_savi_0_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_savi_1_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_savi_2_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_savi_3_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_savi_4_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_videosaur_0_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_videosaur_1_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_videosaur_2_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_videosaur_3_epoch_30_object.ckpt
- cjepa-ckpts/clevrer_videosaur_4_epoch_30_object.ckpt
- cjepa-ckpts/pusht_videosaur_0_epoch_30_object.ckpt
- cjepa-ckpts/pusht_videosaur_1_epoch_30_object.ckpt
- cjepa-ckpts/pusht_videosaur_2_epoch_30_object.ckpt
- clevrer_videosaur_model.ckpt
- pusht_videosaur_model.ckpt
Model artifacts
16 artifacts
Model artifacts
16 artifactscjepa-ckpts/clevrer_savi_0_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 fb74bf9e1daf…4e50 · Hugging Face
Download filecjepa-ckpts/clevrer_savi_1_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 9af6643584e3…9888 · Hugging Face
Download filecjepa-ckpts/clevrer_savi_2_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 5eceff9268c3…21e7 · Hugging Face
Download filecjepa-ckpts/clevrer_savi_3_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 693fac290c6f…ebaf · Hugging Face
Download filecjepa-ckpts/clevrer_savi_4_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 9824af9b2a6d…2d7c · Hugging Face
Download filecjepa-ckpts/clevrer_videosaur_0_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 dfe83d63e7f5…57d5 · Hugging Face
Download filecjepa-ckpts/clevrer_videosaur_1_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 c81363cfadd8…9906 · Hugging Face
Download filecjepa-ckpts/clevrer_videosaur_2_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 c42a751205e3…0c92 · Hugging Face
Download filecjepa-ckpts/clevrer_videosaur_3_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 411ba4624e7c…63fc · Hugging Face
Download filecjepa-ckpts/clevrer_videosaur_4_epoch_30_object.ckpt
ckpt · 13.7 MB · SHA-256 008b084aa08e…8949 · Hugging Face
Download filecjepa-ckpts/pusht_videosaur_0_epoch_30_object.ckpt
ckpt · 101 MB · SHA-256 634c8ac31f71…9dd0 · Hugging Face
Download filecjepa-ckpts/pusht_videosaur_1_epoch_30_object.ckpt
ckpt · 101 MB · SHA-256 8991f56c5cb8…2294 · Hugging Face
Download filecjepa-ckpts/pusht_videosaur_2_epoch_30_object.ckpt
ckpt · 101 MB · SHA-256 d733c9d9cea9…99c8 · Hugging Face
Download fileSource excerpts
3 excerpts
Source excerpts
3 excerpts--- license: apache-2.0 paper_ids: - '2602.11389' pipeline_tag: image-feature-extraction datasets: - clevrer - pusht tags: - object-centric - world-models - causal-inference - jepa - representation-learning - vision --- # C-JEPA: Causal-JEPA This repository contains the weights and code for **Causal-JEPA (C-JEPA)**, a simple and flexible object-centric world model architecture presented in the paper [Causal-JEPA: Learning World Models through Object-Level Latent Interventions](https://huggingface.co/papers/2602.11389). * **Paper:** [Causal-JEPA: Learning World Models through Object-Level Latent Interventions](https://huggingface.co/papers/2602.11389) * **Project Page:** [https://hazel-heejeong-nam.github.io/cjepa/](https://hazel-heejeong-nam.github.io/cjepa/) * **Code:** [https://github.com/galilai-group/cjepa](https://github.com/galilai-group/cjepa) ## Summary World models require robust relational understanding to support prediction, reasoning, and control. While object-centric representations provide a useful abstraction, they are not sufficient to capture interaction-dependent dynamics. C-JEPA is a simple and flexible object-centric world model that extends masked joint embedding prediction from image patches to object-centric representations. By applying object-level masking that requires an object's state to be inferred from other objects, C-JEPA induces latent interventions with counterfactual-like effects and prevents shortcut solutions, making interaction reasoning essential. Empirically, C-JEPA demonstrates: * **Improved Visual Reasoning:** Consistent gains in visual question answering, with an absolute improvement of about 20% in counterfactual reasoning compared to the same architecture without object-level masking on benchmarks like CLEVRER. * **Efficient Planning:** Substantially more efficient planning in agent control tasks (e.g., Push-T), using only 1% of the total latent input features required by patch-based world models while achieving comparable performance. * **Causal Inductive Bias:** A formal analysis demonstrates that object-level masking induces a causal inductive bias via latent interventions. ## Architecture  ## Setup and Usage C-JEPA relies on object-centric encoders (like VideoSAUR or SAVi) to extract representations. For detailed enviro...
Source context: 0 downloads · 9 likes · Pipeline image-feature-extraction · Repo HazelNam/CJEPA
Source context: 0 downloads · 8 likes · Pipeline image-feature-extraction · Repo HazelNam/CJEPA
