genintel commited on
Commit
a86c651
·
verified ·
1 Parent(s): d68f9c0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -6
README.md CHANGED
@@ -15,6 +15,11 @@ size_categories:
15
 
16
  # CRONOS-Benchmark dataset
17
 
 
 
 
 
 
18
  CRONOS-Benchmark is a controlled, synthetic benchmark for evaluating **counterfactual physical consistency** in video world models. It tests whether generative video models correctly simulate three fundamental physical event types — **object falling**, **object collision**, and **object occlusion** — across diverse scenes, objects, and viewpoints.
19
 
20
  Each sequence provides ground-truth RGB frames, depth maps, and segmentation masks, along with a 5-frame conditioning clip and a `metadata.json` file, enabling standardised evaluation of any video generation model that accepts image or video conditioning.
@@ -27,12 +32,6 @@ CRONOS-Benchmark dataset contains **675 sequences** rendered in a controlled sim
27
 
28
  - **License:** Apache 2.0
29
 
30
- ### Dataset Sources
31
-
32
- - **Project Page** [https://genintel.github.io/CRONOS/]
33
- - **Repository:** [https://github.com/GenIntel/CRONOS-benchmark]
34
- - **Paper:** [https://arxiv.org/abs/2605.23699]
35
-
36
 
37
  ## Dataset Structure
38
 
 
15
 
16
  # CRONOS-Benchmark dataset
17
 
18
+ [![Paper](https://img.shields.io/badge/Paper-arXiv-red)](https://arxiv.org/abs/2605.23699)
19
+ [![Project Page](https://img.shields.io/badge/Project-Page-blue)](https://genintel.github.io/CRONOS/)
20
+ [![Dataset](https://img.shields.io/badge/Dataset-HuggingFace-yellow)](https://huggingface.co/datasets/genintel/CRONOS-benchmark)
21
+ [![Code](https://img.shields.io/badge/Code-GitHub-black)](https://github.com/GenIntel/CRONOS-benchmark)
22
+
23
  CRONOS-Benchmark is a controlled, synthetic benchmark for evaluating **counterfactual physical consistency** in video world models. It tests whether generative video models correctly simulate three fundamental physical event types — **object falling**, **object collision**, and **object occlusion** — across diverse scenes, objects, and viewpoints.
24
 
25
  Each sequence provides ground-truth RGB frames, depth maps, and segmentation masks, along with a 5-frame conditioning clip and a `metadata.json` file, enabling standardised evaluation of any video generation model that accepts image or video conditioning.
 
32
 
33
  - **License:** Apache 2.0
34
 
 
 
 
 
 
 
35
 
36
  ## Dataset Structure
37