We introduce CurveBench, a benchmark for testing whether vision-language models can recover hierarchical region-containment trees from images of non-intersecting Jordan curves. The task targets visual topology and structured reasoning beyond simple object recognition, counting, or OCR.</p>\n<p>The Hugging Face collection includes the paper, the CurveBench and CurveBench-Easy datasets, evaluation code, ground-truth generation resources, and fine-tuning artifacts. Our results show that even strong frontier VLMs struggle substantially on the harder settings, while fine-tuned open models improve but remain far from solving the task.</p>\n","updatedAt":"2026-05-15T07:49:03.470Z","author":{"_id":"65e1bdb336a669a4ca5dab7d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65e1bdb336a669a4ca5dab7d/XzxXRFbs6-Sgd8RyoIjYV.png","fullname":"Amir Mohseni","name":"AmirMohseni","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":14,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8417820930480957},"editors":["AmirMohseni"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/65e1bdb336a669a4ca5dab7d/XzxXRFbs6-Sgd8RyoIjYV.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.14068","authors":[{"_id":"6a06c0b6b1a8cbabc9f09a84","name":"Amirreza Mohseni","hidden":false},{"_id":"6a06c0b6b1a8cbabc9f09a85","name":"Mona Mohammadi","hidden":false},{"_id":"6a06c0b6b1a8cbabc9f09a86","name":"Morteza Saghafian","hidden":false},{"_id":"6a06c0b6b1a8cbabc9f09a87","name":"Naser Talebizadeh Saradari","hidden":false}],"publishedAt":"2026-05-13T00:00:00.000Z","submittedOnDailyAt":"2026-05-15T00:00:00.000Z","title":"CurveBench: A Benchmark for Exact Topological Reasoning over Nested Jordan Curves","submittedOnDailyBy":{"_id":"65e1bdb336a669a4ca5dab7d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65e1bdb336a669a4ca5dab7d/XzxXRFbs6-Sgd8RyoIjYV.png","isPro":true,"fullname":"Amir Mohseni","user":"AmirMohseni","type":"user","name":"AmirMohseni"},"summary":"We introduce CurveBench, a benchmark for hierarchical topological reasoning from visual input. CurveBench consists of 756 images of pairwise non-intersecting Jordan curves across easy, polygonal, topographic-inspired, maze-like, and dense counting configurations. Each image is annotated with a rooted tree encoding the containment relations between planar regions. We formulate the task as structured prediction: given an image, a model must recover the full rooted containment tree induced by the curves. Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1 Pro, achieves only 71.1\\% tree-generation accuracy on CurveBench-Easy and 19.1\\% on CurveBench-Hard. We further demonstrate benchmark utility through RLVR-style fine-tuning of open-weight vision-language models. Our trained Qwen3-VL-8B model improves over Qwen-3-VL-8B-Thinking from 2.8\\% to 33.3\\% tree-generation accuracy on CurveBench-Easy, exceeding GPT-5.4 and Claude Opus 4.5 under our evaluation protocol. The remaining gap, especially on CurveBench-Hard, shows that exact topology-aware visual reasoning remains far from solved.","upvotes":0,"discussionId":"6a06c0b6b1a8cbabc9f09a88","projectPage":"https://huggingface.co/collections/AmirMohseni/curvebench","githubRepo":"https://github.com/Amir-Mohseni/CurveBench","githubRepoAddedBy":"user","ai_summary":"CurveBench presents a benchmark for hierarchical topological reasoning using visual inputs, demonstrating significant challenges in exact topology-aware visual reasoning even with advanced models.","ai_keywords":["hierarchical topological reasoning","structured prediction","rooted tree","containment relations","vision-language models","fine-tuning","visual reasoning"],"githubStars":0},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[],"acceptLanguages":["en"],"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.14068.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.14068 in a model README.md to link it from this page.

Cite arxiv.org/abs/2605.14068 in a Space README.md to link it from this page.