Add model card and paper metadata

#1
by nielsr HF Staff - opened
Files changed (1) hide show
  1. README.md +37 -0
README.md ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ library_name: transformers
4
+ pipeline_tag: image-text-to-text
5
+ base_model: Qwen/Qwen2.5-VL-7B-Instruct
6
+ ---
7
+
8
+ # Planning with the Views via Scene Self-Exploration
9
+
10
+ This repository contains a model checkpoint presented in the paper [Planning with the Views via Scene Self-Exploration](https://huggingface.co/papers/2605.29563).
11
+
12
+ [**Project Page**](https://viewsuite.github.io) | [**GitHub**](https://github.com/mll-lab-nu/ViewSuite) | [**Paper**](https://viewsuite.github.io/viewsuite_paper.pdf)
13
+
14
+ ## Overview
15
+
16
+ Can VLMs predict how each camera move changes the view, and plan many such moves ahead? This capability, called **view planning**, requires (1) understanding how a single action transforms the view, and (2) composing many such transformations across multi-turn plans to identify a target view.
17
+
18
+ **ViewSuite** is a 3D point-cloud environment and benchmark suite for view planning, built on real ScanNet indoor scenes. It probes view planning through three diagnostic tasks:
19
+ - **Path-to-View (P2V)**: Predict the resulting view from an action sequence.
20
+ - **View-to-Path (V2P)**: Infer the action sequence between two views.
21
+ - **Interactive View Planning (IVP)**: Plan view changes over multiple turns to identify a target view.
22
+
23
+ This model is an optimized version of **Qwen2.5-VL-7B**, trained using an iterative framework that alternates self-exploration with view graph distillation. This approach significantly closes the planning gap found in frontier VLMs, improving performance on interactive view planning tasks.
24
+
25
+ ## Citation
26
+
27
+ If you find ViewSuite or these checkpoints useful in your research, please consider citing:
28
+
29
+ ```bibtex
30
+ @article{wang2026viewsuite,
31
+ title = {Planning with the Views},
32
+ author = {Wang, Kangrui and Li, Linjie and Yang, Zhengyuan and Chen, Shiqi and
33
+ Wang, Zihan and Fei-Fei, Li and Wu, Jiajun and Guibas, Leonidas and
34
+ Wang, Lijuan and Li, Manling},
35
+ year = {2026}
36
+ }
37
+ ```