Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning
Paper • 2510.27606 • Published • 31
instruction stringclasses 4
values | input stringclasses 1
value | output stringclasses 4
values | images listlengths 1 1 | question_id stringclasses 4
values |
|---|---|---|---|---|
The original image <image> has three regions marked as 1, 2, and 3. Consider the content, positional relationships, depths of the three regions and other cues, and sort the depths of the three regions from smallest to largest from the camera, where the format of the answer is 'Smallest-Middle-Largest'. Your output must... | <think>To determine the depths of the regions marked as 1, 2, and 3, I'll analyze their positions and apparent distances based on cues such as overlaps and known object sizes.
1. **Region 1**: This is a light fixture hanging from the ceiling. It's likely closer to the camera compared to other ceiling elements because ... | [
"coldstart_SFT_images/img_18.jpg"
] | SpatialSSRL_coldstart_18 | |
The image has been divided into 4 labeled 0, 1, 2, and 3. One of the four patches is flipped either horizontally or vertically. The patches are given in <image>. Based on visual clues such as continuity, alignment, and context, answer the correct patch that is flipped and the direction the flip, where the format is 'La... | <think>Let's analyze the image patches:
- **Patch 0**: This section has a clear sky.
- **Patch 1**: This section also has a clear sky, and the airplane is visible.
- **Patch 2**: This patch shows the lower section of the scene with houses.
- **Patch 3**: This patch shows the airplane as well.
Now, consider the alignm... | [
"coldstart_SFT_images/img_24.jpg"
] | SpatialSSRL_coldstart_24 | |
I've taken an image <image> and there are two regions marked as 1, and 2 on the image. Assume that there is a camera at position '2' and it's facing to the left of the image. According to the camera, where is the region marker '1'? A. Front B. Left C. Right D. Back. Consider cues such as depth, orientation, and 3D spat... | <think>To determine the relative position of region marker '1' as seen from the camera at position '2' facing to the left, let's consider the orientation and spatial relationships:
1. **Camera Orientation**: The camera at position '2' is facing to the left of the image.
2. **Spatial Arrangement**: The regions '1' and... | [
"coldstart_SFT_images/img_25.jpg"
] | SpatialSSRL_coldstart_25 | |
The image has been divided into 3 shuffled vertical strips labeled 0, 1, and 2. The shuffled patches are given in <image>. Based on visual clues such as continuity, alignment, and context, answer the correct arrangement of the strips to restore the original image, where the format is 'Top-Middle-Bottom'. Your output m... | <think>To solve the arrangement of the strips:
1. **Visual Clues**:
- Strip 0 features a platform with a person walking and part of a train.
- Strip 1 contains the station platform with multiple people and a view of the station building on the left.
- Strip 2 has a clear sky and a structure that connects with... | [
"coldstart_SFT_images/img_39.jpg"
] | SpatialSSRL_coldstart_39 |
Spatial-SSRL-81k
📖Paper| 🏠Github |🤗Spatial-SSRL-7B Model | 🤗Spatial-SSRL-3B Model | 🤗Spatial-SSRL-Qwen3VL-4B Model | 🤗Spatial-SSRL-81k Dataset | 📰Daily Paper
Spatial-SSRL-81k is a training dataset for enhancing spatial understanding in large vision-language models. It contains 81,053 samples of five pretext tasks for self-supervised learning, offering simple, intrinsic supervision that scales RLVR efficiently.
📢 News
- 🚀 [2026/04/05] We have released the training code of Spatial-SSRL.
- 🚀 [2026/02/25] We have released the 🤗Spatial-SSRL-3B Model, initialized from Qwen2.5-VL-3B-Instruct.
- 🚀 [2026/02/21] Our work has been accepted by CVPR 2026.
- 🚀 [2025/11/24] We have released the 🤗Spatial-SSRL-Qwen3VL-4B Model, initialized from Qwen3-VL-4B-Instruct.
- 🚀 [2025/11/03] Now you can try out Spatial-SSRL-7B on 🤗Spatial-SSRL Space.
- 🚀 [2025/11/03] We have released the 🤗Spatial-SSRL-7B Model, and 🤗Spatial-SSRL-81k Dataset.
- 🚀 [2025/11/02] We have released the 🏠Spatial-SSRL Repository.
🌈 Overview
We are thrilled to introduce Spatial-SSRL, a novel self-supervised RL paradigm aimed at enhancing LVLM spatial understanding. By optimizing Qwen2.5-VL-7B with Spatial-SSRL, the model exhibits stronger spatial intelligence across seven spatial understanding benchmarks in both image and video settings.
💡 Highlights
- 🔥 Highly Scalable: Spatial-SSRL uses ordinary raw RGB and RGB-D images instead of richly-annotated public datasets or manual labels for data curation, making it highly scalable.
- 🔥 Cost-effective: Avoiding the need for human labels or API calls for general LVLMs throughout the entire pipeline endows Spatial-SSRL with cost-effectiveness.
- 🔥 Lightweight: Prior approaches for spatial understanding heavily rely on annotation of external tools, incurring inherent errors in training data and additional cost. In constrast, Spatial-SSRL is completely tool-free and can easily be extended to more self-supervised tasks.
- 🔥 Naturally Verifiable: Intrinsic supervisory signals determined by pretext objectives are naturally verifiable, aligning Spatial-SSRL well with the RLVR paradigm.
🖼️ Task examples
🛠️ Usage
You can find all question-answering pairs in spatialssrl.parquet and the images in images.zip. The images are organized in five folders, each corresponding to a 2D or 3D pretext task.
See the formulation of each task in 📖Paper if you are interested.
✒️Citation
If you find this dataset useful, please kindly cite:
@article{liu2025spatial,
title={Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning},
author={Liu, Yuhong and Zhang, Beichen and Zang, Yuhang and Cao, Yuhang and Xing, Long and Dong, Xiaoyi and Duan, Haodong and Lin, Dahua and Wang, Jiaqi},
journal={arXiv preprint arXiv:2510.27606},
year={2025}
}
📄 License
Usage and License Notices: The data and code are intended and licensed for research use only.
- Downloads last month
- 165